In progress – view active session
Conference 13408
Image-Guided Procedures, Robotic Interventions, and Modeling
16 - 20 February 2025 | Town & Country D
Conference Co-Sponsor
All sponsors 
Show conference sponsors +
View Session
SPIE Medical Imaging Awards and Plenary
16 February 2025 • 5:30 PM - 6:30 PM PST | Town & Country B/C
Session Chairs:
Joseph Y. Lo, Carl E. Ravin Advanced Imaging Labs. (United States), Cristian A. Linte, Rochester Institute of Technology (United States)
View Full Details: spie.org/medical-imaging-awards-plenary
5:30 PM - 5:40 PM:
Symposium Chair Welcome and Best Student Paper Award announcement
First-place winner and runner-up of the Robert F. Wagner All-Conference Best Student Paper Award
Sponsored by:
MIPS and SPIE
5:40 PM - 5:45 PM:
New SPIE Fellow acknowledgments
Each year, SPIE promotes Members as new Fellows of the Society. Join us as we recognize colleagues of the medical imaging community who have been selected.
SPIE Harrison H. Barrett Award in Medical Imaging
Presented in recognition of outstanding accomplishments in medical imaging
13408-500
Development and translation of 3D ultrasound-based imaging systems for diagnostic and image-guided interventions
(Plenary Presentation)
16 February 2025 • 5:50 PM - 6:30 PM PST | Town & Country B/C
Show Abstract +
Our research has been focused on developing 3D US scanning devices that overcome the limitations of conventional US imaging methods. We have been developing and fabricating various mechanical external motorized fixtures that move a conventional US probe in specific patterns and used them in systems for image-guided prostate biopsy prostate, prostate and gynecologic brachytherapy, and focal liver tumour ablation. As well, we developed 3D US-based system for point of care diagnostic application such as whole breast imaging, carotid plaque quantification, and hand and knee osteoarthritis.
Our approach allows scanning the desired anatomy in a consistent manner, imaging a large volume, integration of any manufacturer's 2D US probe into our fixtures, and integration of machine learning methods for rapid diagnosis and guidance. This approach provides a means of using US images with any US system with a small additional cost and minimal environmental constraints.
All-Symposium Welcome Reception
16 February 2025 • 6:30 PM - 8:00 PM PST | Martini Lawn
View Full Details: spie.org/mi/welcome-reception
Join your colleagues on the lawn for food and drinks as we welcome each other to SPIE Medical Imaging 2025.
Monday Morning Keynotes
17 February 2025 • 8:20 AM - 10:30 AM PST | Town & Country B/C
Session Chairs:
Ke Li, The Univ. of Texas M.D. Anderson Cancer Ctr. (United States), Mark A. Anastasio, Univ. of Illinois (United States), Shandong Wu, Univ. of Pittsburgh (United States)
View Full Details: spie.org/monday-morning-keynotes
8:20 AM - 8:25 AM:
Welcome and introduction
8:25 AM - 8:30 AM:
Award announcements
- Robert F. Wagner Award finalists for conferences 13405, 13409, and 13411
- Physics of Medical Imaging Student Paper Award
13405-501
Promoting innovation in your team: Lessons learned from 40 years in medical imaging
(Keynote Presentation)
17 February 2025 • 8:30 AM - 9:10 AM PST | Town & Country B/C
Show Abstract +
The development of advanced cross-sectional imaging technologies, especially X-ray CT and MRI, are widely recognized as the most impactful inventions in health care during the last 50 years. During this period of transformative innovation in medical imaging, progress has been accelerated through collaborative efforts between medical physicists, physicians, and the medical imaging industry. Innovation can be accelerated through individual efforts to promote the creative process, as well as frameworks to enhance collaboration and invention amongst teams of researchers. The purpose of this lecture is to examine key elements of the inventive process that have contributed to the development of medical imaging in the past that can be leveraged for ongoing advances in healthcare in the future.
13409-502
Designing AI for the clinic: Internalizing the clinical task when developing deep-learning algorithms for medical imaging
(Keynote Presentation)
17 February 2025 • 9:10 AM - 9:50 AM PST | Town & Country B/C
Show Abstract +
Deep learning algorithms for image reconstruction, processing and segmentation are being developed and showing strong promise for multiple medical-imaging applications. However, medical images are acquired for clinical tasks, such as defect detection and feature quantification, and these algorithms are typically developed and evaluated agnostic to this clinical task. This talk will first discuss why clinical-task-agnostic evaluation of AI algorithms can be misleading, emphasizing the need for clinical-task-based evaluation of these algorithms. This discussion will lead to a new frontier in designing deep-learning algorithms that explicitly account for the clinical task of interest. We will see through examples how the clinical task can be incorporated within the design of these algorithms and how this then poises the algorithm for success in clinical applications. Throughout the talk, we will note how the rich field of model observers provides a mechanism to both design and evaluate AI algorithms for clinical tasks.
13411-503
To be determined
(Keynote Presentation)
17 February 2025 • 9:50 AM - 10:30 AM PST | Town & Country B/C
Break
Coffee Break 10:30 AM - 11:00 AM
Session 1:
Surgical Navigation
17 February 2025 • 11:00 AM - 12:40 PM PST | Town & Country D
Session Chairs:
William E. Higgins, The Pennsylvania State Univ. (United States), Baowei Fei, The Univ. of Texas at Dallas (United States)
13408-1
17 February 2025 • 11:00 AM - 11:20 AM PST | Town & Country D
Show Abstract +
This work presents a real-time, learning-based solution for registering MR to intraoperative long-length tomosynthesis imaging to resolve anatomical deformations and facilitate MR-based navigation in spine surgery. A 3D-2D pose regression network was developed to predict the 3D poses of individual vertebrae by leveraging the calibrated system geometry and features extracted from soft-tissue-suppressed projection images. The results show a median TRE of 0.9 mm in CT-to-LF and 2.3 mm in MR-to-LF registration. With an average runtime of <1 s, the approach achieved a 200× speedup over a conventional optimization-based approach, demonstrating its potential for rapid pose corrections during spine surgery.
13408-2
17 February 2025 • 11:20 AM - 11:40 AM PST | Town & Country D
Show Abstract +
Background: Surgical navigation is growing to standard of care in spine surgery. However, any spinal motion occurring after volumetric reference imaging deteriorates registration accuracy. A level-wise, stereovision based image updating pipeline has been developed to account for this motion.
Objective: We aim to automate the detection of the spine exposure in stereovision images by utilizing the Segment Anything Model 2 (SAM2) thereby automating the input to the level-wise registration pipeline.
Methods: We obtained preoperative CT, intraoperative CT, and tracked stereovision images of a cadaver swine’s spine. SAM2 automatically segments the exposure in an initial stereovision image and propogates the mask across images. Depth maps are used to refine the masks creating an automatic mask generation to inform the input to a level-wise registration pipeline.
Results: The automated masks perform comparably to a manually segmented mask for level-wise registration accuracy.
Conclusions: The automated method of spine exposure detection in stereovision aids to remove user dependance from the input to our level-wise registration pipeline in open spine surgery.
13408-3
17 February 2025 • 11:40 AM - 12:00 PM PST | Town & Country D
Show Abstract +
Traditional cochlear implant (CI) surgery often lacks tool tracking and detailed preoperative planning, which can result in suboptimal implant placement. This study introduces a novel computer vision-based method to enhance CI surgery through augmented reality. The method leverages Meta's Segment Anything Model 2 (SAM 2) for tool segmentation and employs a 3D-to-2D registration technique using gradient descent optimization and differentiable rendering for real-time pose estimation. Validation on synthetic and real surgical datasets demonstrates a mean absolute rotation error of less than 1 degree and a translation error of less than 1 mm, highlighting its potential for improving the accuracy and outcomes of CI surgeries.
13408-4
17 February 2025 • 12:00 PM - 12:20 PM PST | Town & Country D
Show Abstract +
In our paper, we introduce a novel pipeline that is capable of generating synthetic multi-view videos from a single CI microscope image given their corresponding CT mesh. The microscope multi-view synthesis holds promise for various applications, particularly in intraoperative registration and robot-assisted surgery. Our goal is to create realistic visualizations that can assist intraoperative navigation in the Operating Room (OR) and potentially improve the outcomes of CI surgeries. Furthermore, the proposed methodology addresses a critical challenge in the medical domain: the shortage of surgical videos for training deep learning-based models. By requiring only a single microscope view and its corresponding patient’s CT scan, our approach can generate thousands of realistic visualizations. This feature is particularly helpful in CI surgeries, where extensive frame-by-frame pose annotation can be replaced with synthetic views generation.
13408-5
17 February 2025 • 12:20 PM - 12:40 PM PST | Town & Country D
Show Abstract +
This paper presents an integrated surgical navigation system for liver surgery, fully implemented within the HoloLens® 2 platform. The system seamlessly combines real-time tool tracking, deformation correction, and dynamic 3D model visualization in a mixed reality environment, all within a single, cohesive workflow. The system’s tracking accuracy was validated through both point and surface measurements, showcasing its potential to enhance intraoperative guidance. By leveraging the comprehensive capabilities of HoloLens® 2, this work introduces a streamlined, efficient approach to surgical navigation, aimed at improving precision and outcomes in liver surgery. Future research will extend its application to subsurface targets to evaluate the accuracy of the system.
Break
Lunch Break 12:40 PM - 1:40 PM
Session 2:
Interventional Radiology and Minimally Invasive Surgery
17 February 2025 • 1:40 PM - 3:00 PM PST | Town & Country D
Session Chairs:
Matthieu Chabanas, Univ. Grenoble Alpes (France), Elvis C.S. Chen, Robarts Research Institute (Canada)
13408-6
17 February 2025 • 1:40 PM - 2:00 PM PST | Town & Country D
Show Abstract +
Breast cancer remains the most common cancer among women worldwide. Breast needle biopsy, guided primarily by ultrasound, is the primary way to definitively confirm a breast cancer diagnosis. However, it exhibits a high rate of false-negative cases, which can be attributed to operator-dependence. For women with dense breast tissue where lesions are not visualized under ultrasound, MRI-guided breast needle biopsies are recommended instead. However, the MRI-based procedure can be inaccessible and time-consuming. To address these issues, we have modified our previously developed 3D automated breast ultrasound (3D ABUS) system, compatible with any commercial ultrasound probe to accommodate breast needle biopsy procedures. This work represents a notable advancement in integrating 3D ABUS with breast needle biopsy, offering a more cost-effective and accessible solution to the procedure. Overall, our system demonstrates the potential to improve breast biopsy procedure workflow and patient experience, particularly in women with dense breasts.
13408-7
17 February 2025 • 2:00 PM - 2:20 PM PST | Town & Country D
Show Abstract +
Implementation of MRI guidance in place of CT or X-ray would reduce ionizing radiation exposure and provide better soft tissue contrast in a wide range of musculoskeletal targets in children. We developed an augmented reality guidance system for MR-guided interventions. The system allowed interventional radiologist to plan a needle trajectory and projected guidance information onto the skin through an optical head. The interventional radiologist advanced the needle while following the real-time guidance to reach the desired target. The purpose of this study was to evaluate feasibility of using this system to target shoulder, hip, sacroiliac joints and liver in a cadaver. In total, 12 locations were targeted, and all needles were successfully placed. The system was subjectively judged to be easy to use and to fit within the established clinical workflow of image-guided needle-based interventions.
13408-8
17 February 2025 • 2:20 PM - 2:40 PM PST | Town & Country D
Show Abstract +
In this paper, we propose to automatically segment Alexis retractors on video endoscopic images recorded during a video-guided thoracic surgery (VATS), and on an intraoperative CBCT acquired during this surgery. On endoscopic video images, Alexis can be segmented by simple hue thresholding. On CBCT, as there is no contrast between the tissue and the Alexis, we had to characterize the Alexis retractors by the curvature of their surfaces. We have set up a processing framework that automatically extracts the internal surface of the rib-cage, characterizes the boundaries of the Alexis retractors by their surface mean curvature, separates the different Alexis and models them as elliptical rings. In a future study, the features segmented in this two modalities will be used for registration purpose, enabling augmented reality during VATS.
13408-9
17 February 2025 • 2:40 PM - 3:00 PM PST | Town & Country D
Show Abstract +
Minimally invasive surgery (MIS) has benefitted greatly from advancements such as fluorescence and narrow-band imaging. Modern computer vision (CV) architectures and foundation models have shown promise in classification tasks and may be used in medical imaging systems, but these imaging modalities have drawbacks hindering the ability to generate the large datasets required of modern CV models. Hyperspectral imaging (HSI) is non-invasive and contrast-free and has shown promise in tissue classification and cancer detection. But, to leverage the rich information provided with HSI alongside modern CV architectures, larger datasets must be curated. As such, we have designed an HSI system and workflow to address this gap by allowing for high-throughput ex vivo tissue imaging. The system has been used to acquire 373 hyperspectral images of tissues from four animal models including porcine, murine, galline, and bovine organs to date.
Break
Coffee Break 3:00 PM - 3:30 PM
Session 3:
Video-based Interventional Applications
17 February 2025 • 3:30 PM - 5:30 PM PST | Town & Country D
Session Chairs:
Kristy K. Brock, The Univ. of Texas M.D. Anderson Cancer Ctr. (United States), Jack H. Noble, Vanderbilt Univ. (United States)
13408-10
17 February 2025 • 3:30 PM - 3:50 PM PST | Town & Country D
Show Abstract +
Monocular depth estimation is crucial for tracking and reconstruction algorithms, particularly in the context of surgical videos. However, the inherent challenges in directly obtaining ground truth depth maps during surgery render supervised learning approaches impractical. While many self-supervised methods based on Structure from Motion (SfM) have shown promising results, they rely heavily on high-quality camera motion and require optimization on a per-patient basis. These limitations can be mitigated by leveraging the current state-of-the-art foundational model for depth estimation, Depth Anything. However, when directly applied to surgical scenes, Depth Anything struggles with issues such as blurring, bleeding, and reflections, resulting in suboptimal performance. This paper presents a fine-tuning of the Depth Anything model specifically for the surgical domain, aiming to deliver more accurate pixel-wise depth maps tailored to the unique requirements and challenges of surgical environments.
13408-11
17 February 2025 • 3:50 PM - 4:10 PM PST | Town & Country D
Show Abstract +
While NeRF has been extensively applied to generate implicit, qualitative reconstructions in vivo, it has yet to be effectively implemented and extensively tested in the quantitative representation of anatomical geometry. Our expanded workflow demonstrates the ability to create high-resolution and accurate 3D reconstructions of surgical scene anatomy. Using a series of three cadaveric specimens, measurements of critical anatomy were reconstructed and evaluated with average errors for ethmoid length and height being 0.17mm and 0.70mm, respectively.
13408-12
17 February 2025 • 4:10 PM - 4:30 PM PST | Town & Country D
Show Abstract +
Monocular depth estimation from kidney endoscopy requires ground truth depth information for endoscopy videos in order to train a supervised neural network. Kidney endoscopy is not compatible with real-time tracking of the scope tip via, e.g., optical trackers, which makes it difficult to directly obtain such depth information. To address this problem, we propose a novel pipeline to reconstruct 3D point cloud from the endoscopy video, which we register to a segmentation obtained from the preoperative CT image. We then use the estimated camera poses from the reconstruction and the segmented preoperative CT to create synthetic endoscopy frames and accurate depth information. We validate this method on kidney phantoms and real patient data.
13408-13
17 February 2025 • 4:30 PM - 4:50 PM PST | Town & Country D
Show Abstract +
Fully supervised deep learning (DL) models for surgical video segmentation have been shown to struggle with non-adversarial, real-world corruptions of image quality including smoke, bleeding, and low illumination. Foundation models for image segmentation, such as the segment anything model (SAM) that focuses on interactive prompt-based segmentation, move away from semantic classes and thus can be trained on larger and more diverse data, which offers outstanding zero-shot generalization with appropriate user prompts. Recently, building upon this success, SAM-2 has been proposed to further extend the zero-shot interactive segmentation capabilities from independent frame-by-frame to video segmentation. In this paper, we present a first experimental study evaluating SAM-2's performance on surgical video data. Leveraging the SegSTRONG-C MICCAI EndoVIS 2024 sub-challenge dataset, we assess SAM-2's effectiveness on uncorrupted endoscopic sequences and evaluate its non-adversarial robustness on videos with corrupted image quality simulating smoke, bleeding, and low brightness conditions under various prompt strategies.
13408-14
17 February 2025 • 4:50 PM - 5:10 PM PST | Town & Country D
Show Abstract +
Video endoscopy has significantly advanced gastrointestinal disease investigation, but reviewing endoscopy videos often requires frequent adjustments and reorientations, making the process time-consuming and error-prone. Image stitching offers a solution by providing a continuous view of the examined area, yet esophageal images present challenges due to their smooth surfaces, lack of distinct features, and non-horizontal orientation, making traditional methods less effective. To address this, we propose a novel preprocessing pipeline that enhances endoscopic image stitching. Our approach converts video data into continuous 2D images through four steps: (1) keyframe selection, (2) rotation correction, (3) surface unwrapping using polar coordinates, and (4) feature point matching enhanced by Adaptive Histogram Equalization. Testing on 20 pediatric endoscopy videos shows significant improvements in image alignment and stitching quality, establishing a strong foundation for panoramic image creation.
13408-15
Automated segmentation of central airway obstruction from endoscopic video stream with deep learning
17 February 2025 • 5:10 PM - 5:30 PM PST | Town & Country D
Show Abstract +
Central airway obstruction (CAO) can disrupt normal breathing and pose significant risks, with treatments often involving surgical removal.
Accurate image segmentation is crucial for identifying the target region to assist surgeons or robotic surgery system during operations. Although manual annotation is the gold standard, it is time-consuming and subjective, making a robust automated segmentation algorithm desirable.
In this paper, we propose an automated deep learning framework for CAO segmentation. As a benchmark study, we build a custom CAO phantom model and acquire endoscopy videos. We then establish inter-rater variability.
To assess the effectiveness of the proposed method, a 4-fold cross-validation is performed with The Dice score as an evaluation metric. The proposed CAO segmentation method yields mean binary Dice scores of 0.95 and 0.89 against annotations from two human raters, respectively, compared to an inter-rater variability Dice score of 0.88. This indicates that the proposed framework can provide robust CAO segmentation from endoscopic video.
Posters - Monday
17 February 2025 • 5:30 PM - 7:00 PM PST | Golden State Ballroom
View Full Details: spie.org/mi/monday-poster-session
Conference attendees are invited to attend the SPIE Medical Imaging poster session on Monday evening. Come view the posters, enjoy light refreshments, ask questions, and network with colleagues in your field. Authors of poster papers will be present to answer questions concerning their papers. Attendees are required to wear their conference registration badges.
Poster Presenters:
Poster Setup and Pre-Session Viewing: 10:00 AM - 5:30 PM Monday
- In order to be considered for a poster award, it is recommended to have your poster set up by 1:00 PM Monday. Judging may begin after this time. Posters must remain on display until the end of the Monday evening poster session, but may be left hanging until 1:00 PM Tuesday. After 1:00 PM on Tuesday, posters will be removed and discarded.
https://spie.org/MI/Poster-Presentation-Guidelines
13408-45
17 February 2025 • 5:30 PM - 7:00 PM PST | Golden State Ballroom
Show Abstract +
Fiber Optic Shape Sensing (FOSS) is a technology that visualizes the three-dimensional shape of medical devices using an optical fiber embedded in the device. This technology enhances the precision and safety of medical interventions by providing real-time, accurate visualization of the device's position within the body. The proposed study addresses challenges related to FOSS technology, such as data compression and noise. By using AI-powered compressed sensing and sparsifying transforms, the research aims to reduce FOSS data streams, retaining only relevant information for downstream applications. This efficient data handling minimizes storage and transmission requirements and ensures the retrieval of actionable insights. The integration of AI with FOSS technology promises to improve medical device navigation accuracy and patient outcomes, potentially leading to widespread adoption in minimally invasive procedures.
13408-46
17 February 2025 • 5:30 PM - 7:00 PM PST | Golden State Ballroom
Show Abstract +
Prostate cancer (PCa) remains a significant public health challenge worldwide, as it is the most diagnosed cancer among men. Traditional 2D transrectal ultrasound (TRUS)-guided biopsies have high false-negative rates, necessitating repeat procedures. The novel prostate-specific PET (P-PET) system developed by Radialis offers increased sensitivity and resolution, addressing the limitations of whole-body PET systems. This study aims to develop and clinically validate an integrated 3D TRUS and P-PET-guided prostate biopsy system, incorporating robotic technologies for precise needle guidance. Tests confirm the system’s potential to enhance PCa biopsy accuracy by reducing false-negative rates and minimizing the need for repeat biopsies. Future steps include co-registering ultrasound and P-PET images, quantifying needle guidance errors, and validating the system’s effectiveness in clinical trials.
13408-47
17 February 2025 • 5:30 PM - 7:00 PM PST | Golden State Ballroom
Show Abstract +
Fluoroscopically-guided interventions (FGIs) offer an interesting link between medical imaging and more traditional procedural interventions such as surgery. This work outlines the initial development and application of methods to capture and analyze data streams in FGI procedures. Data sources included four distinct but temporally aligned video feeds from each FGI procedure room, radiation dose structured reports (RDSR), and electronic medical record (EMR) data. Automated pipelines succeeded in converting procedure videos into sequential image frames at several different frame rates. Simple image mask and intensity analysis enabled consistent categorization of each frame by radiation event type. A more complete analysis of FGI procedures was possible by aligning RDSR data with image frames from the actual procedure. Key fingerprints of critical steps in FGI procedures could be identified by comparing changes in image acquisition parameters from the RDSR with actual frames from procedure videos for verification. This data capture and analysis framework offers many opportunities for automation to enhance efficiency and improve FGI workflows and procedures.
13408-48
17 February 2025 • 5:30 PM - 7:00 PM PST | Golden State Ballroom
Show Abstract +
Many spinal operations are performed using fluoroscopic guidance; however, its efficacy is conditional on accurate needle placement. Image-guided surgical navigation systems allow for intraoperative localization of surgical tools, significantly improving patient outcomes. Electromagnetic navigation systems require a field generator (FG) whose placement must be near the patient and may partially obstruct the x-ray beam, potentially degrading image quality due to x-ray scatter interactions from the FG. This may reduce image contrast, add noise and decrease spatial resolution. These scatter interactions can be assessed in terms of the scatter-to-primary ratio and its effect on image quality can be described using the modulation transfer function and the detective quantum efficiency. This work aims to quantify the effect of the FG on image quality using the aforementioned parameters to ensure minimal deterioration, with the goal to move towards the seamless integration of EM tracking for fluoroscopy-guided interventions.
13408-49
17 February 2025 • 5:30 PM - 7:00 PM PST | Golden State Ballroom
Show Abstract +
Accurate alignment of pre-operative planning data with intra-operative conditions is critical for effective surgical navigation. In neurosurgery, this alignment is complicated by brain-shift, which occurs when the brain shifts within the skull after the dura mater is opened and cerebrospinal fluid is drained. We propose a novel method for intra-operative alignment of MRI-based planning data with photographs taken during neurosurgery, using a landmark-driven 2D/3D registration technique. Landmarks are interactively selected on both the intra-operative photographs and 3D renderings of the brain’s primary vasculature, derived from T1-weighted MR images with and without contrast enhancement. Registration is performed by minimizing the average distance between corresponding landmark pairs in the two modalities. The accuracy of this method is evaluated by comparing the positions of physically placed markers on the brain surface across multiple photographs taken from different angles.
13408-50
17 February 2025 • 5:30 PM - 7:00 PM PST | Golden State Ballroom
Show Abstract +
Guided bronchoscopy systems are revolutionizing bronchoscopy. These systems draw on a procedure plan derived from a 3D chest CT scan consisting of an airway route leading to an airway close to a region of interest. Unfortunately, the virtual chest space, captured by the planning CT, differs significantly from the real chest space arising during the live procedure. This CT-to-body divergence (CTBD) can produce large guidance errors. CTBD arises because the procedure plan relies on a CT scan depicting the chest near total lung capacity (TLC), while the live procedure occurs with the chest near functional residual capacity (FRC). These volume differences lead to mismatches between the guidance system's perceived 3D position and bronchoscope's actual position. We propose a method to help mitigate CTBD on guided bronchoscopy by using spatial information from CT scans obtained at two lung states. Phase one entails a procedure plan giving two point-by-point synchronized procedure plans: one in TLC space and the other in FRC space. Phase two involves a guidance method giving synchronized information in both TLC and FRC spaces. Results from TLC/FRC CT pairs verify the method’s potential.
13408-51
17 February 2025 • 5:30 PM - 7:00 PM PST | Golden State Ballroom
Show Abstract +
A new VSLAM framework is proposed to combine the strengths of traditional SLAM algorithms and LoFTR. This framework is designed to enable robust and accurate tracking of bronchoscopic video sequences, even in challenging environments such as bronchoscopy procedures. By integrating LoFTR into the SLAM pipeline, our framework can handle such environments with high accuracy and reliability. Additionally, our VSLAM framework incorporates monocular depth estimation to generate more 3D points. This seamless integration of LoFTR and depth estimation allows for the tracking of longer video sequences.
13408-52
17 February 2025 • 5:30 PM - 7:00 PM PST | Golden State Ballroom
Show Abstract +
Precise tumor targeting and temperature monitoring is necessary during thermal ablation of liver cancer. Ultrasound (U/S) is used for targeting due to its real-time capabilities, while it lacks tumor conspicuity that MRI/CT offers. There is an unmet need in combining interventional U/S with pre-interventional MRI as well as a need for MR thermometry to monitor tumor temperature to avoid undertreating the tumor and overtreating the surrounding healthy tissues. Therefore, we present a procedural ablation workflow that can stream U/S, deformably register 3D U/S and MR, and display both deformed MR landmarks and MR thermometry all in near real-time.
13408-53
17 February 2025 • 5:30 PM - 7:00 PM PST | Golden State Ballroom
Show Abstract +
Background: Real-time tracking of regions of interest is crucial in clinical procedures, yet many systems are limited to 2D tracking and suffer from delays in image acquisition and processing. This research introduces methods to estimate 3D target movements in real time using interleaved coronal and sagittal MR images.
Methods: Image registration is employed to measure 2D target movement from a reference image. This 2D information is then combined with predictions from LightGBM models to determine the entire 3D displacement of the target. Additionally, the 2D measurements are incorporated into the LightGBM training set for continuous online re-optimization. The methods were evaluated using a curated dataset of real liver data.
Results: On average, the image registration method yielded tracking errors of 1.19 mm. System delays of 200 ms, 400 ms and 600 ms} led to tracking errors of 1.78 mm, 2.62 mm and 3.46 mm. The trained LightGBM models, once trained, reduced these errors by 29% to 46%.
Conclusions: Our methods effectively address data gaps and system delays, showing promise for real-time 3D tracking of targets without needing prior displacement knowledge.
13408-54
17 February 2025 • 5:30 PM - 7:00 PM PST | Golden State Ballroom
Show Abstract +
Liver surgery faces challenges, primarily due to the limited intra-operative visibility of the liver surface. Many registration methods have been developed to assist clinicians to better target regions of interest during a surgical procedure. The registration process to align the partial intra-operative surface with the pre-operative surface struggles due to partial visibility. Addressing this challenge is vital for precise alignment of liver surfaces. Learning-based point cloud completion methods have the potential to alleviate this issue. Thus, we explore six state-of-the-art point cloud completion methods to identify the optimal method, focusing on a patient-specific approach in two cases: canonical and non-canonical pose. Transformer-based methods, PoinTr and AdaPoinTr, outperform all other methods to generate a complete point cloud from the given partial liver point cloud for the canonical pose. The promising results shown by these techniques may enhance accurate registration of pre- and intra-operative liver surfaces, improving overall surgical outcome.
13408-55
17 February 2025 • 5:30 PM - 7:00 PM PST | Golden State Ballroom
Show Abstract +
Ultrasound (US) image-guided access is widely used in surgical operations. One major challenge conventional US image-guided access faces is sustaining proper alignment between the needle path and the US image plane to maximize needle visibility, since the needle and the US transducer are controlled by each individual hand. Reflector-integrated ultrasound (ref-US) image-guided access was originally proposed to provide by-default alignment of the needle path and the image plane, however issues like bulky size, unconventional handling style and encapsulation of acoustic medium pose difficulties to clinical application of ref-US image-guided access. In this work we focused on improving clinical applicability of ref-US image-guided needle access mechanism. First, the size of the ref-US image guided access mechanism is minimized, and handling style of the mechanism is modified to follow the handling style of a normal ultrasound transducer by using double acoustic reflectors. we use gel-like medium to encapsulate ref-US image-guided access mechanism to reduce imaging artifacts and to prevent medium leakage, while both are common issues in encapsulation using liquid medium.
13408-56
17 February 2025 • 5:30 PM - 7:00 PM PST | Golden State Ballroom
Show Abstract +
The feasibility of monitoring percutaneous microwave thermal ablation (MWA) using ultrasound (US) echo decorrelation imaging in patients with hepatocellular carcinoma (HCC) (N=7) was tested in this paper. MWA was performed using standard-of-care MWA generator and was guided using 128 element US arrays with imaging frequency 2.2 MHz. To monitor ablation progress, 20 beamformed RF frames were acquired throughout the ablation process until the treatment ceased automatically due to elevation of tissue impedance or exceeding predetermined maximum number of treatment cycles. For comparison, echo decorrelation imaging and integrated backscattered (IBS) images were computed for each patient and co-registered with corresponding post-procedure computed tomography images. Both imaging methods were tested using receiver operating characteristic (ROC) curve. Area under the ROC (AUC) were computed to compare both methods statistically. Results showed that echo decorrelation imaging predicted percutaneous MWA in patients with HCC significantly better than IBS (AUC = 0.772, AUC = 0.675, p=0.0027, respectively).
13408-57
17 February 2025 • 5:30 PM - 7:00 PM PST | Golden State Ballroom
Show Abstract +
The aim of this study is to enhance prostate surgery visualization for real-time prostate tracking by integrating AR with 3D MR and PET volume data. PET/MR images were obtained 3D anatomical and functional information. A 10-second surgical video part that contain various situations was selected. The prostate was manually segmented from MR images using 3D Slicer software to create a binary mask. This mask was applied to extract prostate volume data from the registered PET/MR images. The fast and robust MOSSE method was employed for tracking the prostate in the surgical video. The tracked prostate volume data was visualized using a ray-casting technique, creating a realistic 3D volume data visualization. Results showed that our model effectively tracked the prostate from the surgical video despite varying conditions. SSIM evaluation showed that ray-casted PET images had a 4.14% higher similarity to the tracked regions compared to MRI images. These results demonstrate that our model can be used to enhance the outcome of prostate surgery by providing real-time MR anatomical and PET functional information.
13408-58
17 February 2025 • 5:30 PM - 7:00 PM PST | Golden State Ballroom
Show Abstract +
Intraoperative guidance aids in placement of pedicle screws during operation but can still lead to misplacement, therefore breach, as reported from anywhere between 3% to 13% with minimally invasive operations and 2% to 35% for open screw placement. This study defines an algorithm for breach prevention during orthopedic drilling operations using a suite of sensory inputs including but not limited to distance, force, speed, vibration, and audio. An off the shelf drill press was retrofitted with 3D printed fixturing, the sensors, and was automated for a constant linear federate with a linear actuator. A Stryker System 6 drill was attached to the system and operated in a continuous rotational state at its maximum speed. The specimen used in the study are MDF, Plywood, Wood Stack (MDF+Plywood), and porcine bone. The results signify that breach can be detected preemptively through a sensor feedback combination algorithm and halted with 4.25mm remaining.
13408-59
17 February 2025 • 5:30 PM - 7:00 PM PST | Golden State Ballroom
Show Abstract +
In image-guided neurosurgery, post-dural opening factors degrade the accuracy of pre-operative magnetic resonance (pMR) images due to brain deformation. Our approach updates pMR images to match the surgical scene, using a navigation system and an intraoperative stereovision (iSV) system mounted on a surgical microscope. Recently, exoscopes like the Synaptive Modus X, which integrate navigation and 3D visualization, have gained popularity. This study integrates stereovision data from an exoscope into our image updating pipeline, adapting iSV acquisition and reconstruction. Navigation data were acquired via OpenIGTLink, and image data via frame grabbers. Stereo parameters and spatial relationships were calibrated using a tracked checkerboard. Results showed comparable accuracy between exoscope-based and microscope-based systems, with both achieving high precision in camera calibration and checkerboard point reconstruction. The exoscope-based system effectively provides intraoperative data for brain deformation compensation.
13408-60
17 February 2025 • 5:30 PM - 7:00 PM PST | Golden State Ballroom
Show Abstract +
A novel video generation pipeline is proposed to create datasets of tool-bowel interaction videos for robust training of haptic assistance models. To evaluate the pipeline, modified models built on convolution-based architectures are implemented to train and test on the artificial videos. Results demonstrate a high level of flexibility and customizability in the pipeline with varied levels of success of the models. Limitations of existing models to learn in complex environmental settings are demonstrated, indicating the proposed pipeline could be an effective tool for generating datasets for robust training of force estimation models.
13408-61
17 February 2025 • 5:30 PM - 7:00 PM PST | Golden State Ballroom
Show Abstract +
Automatic analysis of surgical activities is an imperative for providing context-aware decision support in operating rooms. While fine-grained detections have been proposed to enhance surgical understanding, the growing refinement leads to an increased difficulty in the task. This study aims to improve the recognition performance of surgical scenes by introducing a coarse-to-fine strategy. The initial phase involves generating robust coarse outputs through an ensemble of multi-task recurrent neural networks (RCNN), specifically targeting phase, tool, action, and target recognition. Subsequently, these outputs are employed to learn the triplet label, facilitating a detailed understanding of surgical scenes. Thus, the triplet detection is estimated from an initial coarse detection, as opposed to adopting an end-to-end methodology for learning the triplet label. Experimental results demonstrate an overall improvement across all tasks when compared to individual networks. Furthermore, the strategy showed a promising performance in the CholecTriplet2021 challenge, with an Average Precision of 36.9 comparable to the state-of-the-art, proving its added value to the community.
13408-62
17 February 2025 • 5:30 PM - 7:00 PM PST | Golden State Ballroom
Show Abstract +
This paper presents a method for estimating hidden blood vessel locations in laparoscopic videos by combining preoperative and intraoperative information. Blood vessels are often hidden by adipose tissues and other anatomical structures. A computer-aided surgery system that reveals the locations of these concealed blood vessels could assist surgeons. Therefore, we develop a method for estimating the hidden blood vessels in laparoscopic videos. The proposed method estimates the location of the hidden blood vessels by using both the preoperative information and the intraoperative laparoscopic videos. We apply fully convolutional networks (FCNs) to the laparoscopic video frames to estimate these locations. Our FCN model utilizes attention mechanisms to incorporate the preoperative information. The preoperative information is computed from preoperative CT volumes at the first frame and is deformed using the optical flow in the following frames. The experimental results showed that the proposed method could estimate the locations of the hidden blood vessel in laparoscopic videos.
13408-63
17 February 2025 • 5:30 PM - 7:00 PM PST | Golden State Ballroom
Show Abstract +
Laparoscopic surgery is a widely used minimally invasive procedure to treat internal organs mostly in abdominal or pelvic area. However, surgeons have limited field of view during operation because surgical instruments are inserted through patients’ body via small incisions. To improve surgeon’s perception, surgical navigation can help improve laparoscopic surgery. Learning-based computer vision algorithms and neural radiance field (NeRF) models can be applied to enhance surgical navigation system's efficiency. In this work, we generate novel views, depths, and camera poses of corresponding laparoscopic images from an existing dataset to augment our dataset. We performed tissue surface reconstruction of the dataset via NeRF-SLAM algorithm. Camera poses were transformed to acquire novel views of RGB laparoscopic images and their corresponding depth images. The results show that the simulated RGB images have peak signal-to-noise ratios (PSNR) between 35.47 – 36.51 dB compared to the corresponding input images. This work provides a foundation for the generation of synthetic datasets which could be useful for learning-based autonomous surgical navigation systems.
13408-64
17 February 2025 • 5:30 PM - 7:00 PM PST | Golden State Ballroom
Show Abstract +
In healthcare, the laparoscopic prostatectomy procedure is used when less invasive treatments are ineffective or unavailable. While laparoscopic prostatectomy can be highly effective, it comes with challenges such as complexity, the risks of complications, and other post-operative issues. Our approach uses multiple generative artificial intelligence models to provide surgeons with decision-making assistance while taking into account the interaction medium. The proposed surgery copilot was tested in a private dataset that is composed of 250 laparoscopic prostatectomy questions. The preliminary results obtained suggested the methodology implemented has the potential to be integrated into a surgery room to clarify the surgeon about laparoscopic prostatectomy.
13408-65
17 February 2025 • 5:30 PM - 7:00 PM PST | Golden State Ballroom
Show Abstract +
This project seeks to advance the data acquisition and processing capabilities of a dedicated head and neck PET system by integrating a machine learning model onto an FPGA using the HLS4ML library. This model detects multiple photon interactions within a Cadmium Zinc Telluride crystal, enhancing the accuracy and resolution of PET scans. By converting high-level machine learning models from TensorFlow and PyTorch into FPGA-compatible code, the system facilitates real-time data processing. We propose replacing the traditional ADC with a 1-bit Delta-sigma ADC that simplifies the data pipeline and increases processing efficiency, ultimately speeding up image reconstruction and improving diagnostic precision.
13408-66
17 February 2025 • 5:30 PM - 7:00 PM PST | Golden State Ballroom
Show Abstract +
Minimally invasive surgery (MIS) aims to replace traditional open surgery, reducing blood loss and increasing patient comfort through small incisions for instrument and camera placement. Currently, robotic systems controlled by surgeons hold the camera, but automation is needed.
This work presents a new proof of concept for dynamically controlling a laparoscopic camera using robotic systems. Tracking experiments and the robot's ability to follow the information obtained in real-time from new positions are demonstrated.
13408-67
17 February 2025 • 5:30 PM - 7:00 PM PST | Golden State Ballroom
Show Abstract +
Cardiovascular diseases are currently the leading cause of death globally. These issues are associated with unhealthy diets high in sugar and saturated fats, combined with stressful careers and sedentary lifestyles, leading to the accumulation of fat around various organs. The treatment for these conditions usually involves minimally-invasive procedures guided by medical imaging, with ultrasound (US) imaging being a preferred radiation-free modality. However, the US requires experienced radiologists, particularly for manipulating a transesophageal probe. This work aims to develop a robot to handle a transesophageal probe to ease US-guided cardiovascular treatments. To automatically adjust the probe's position towards the anatomical target, a motion-tracking system was attached to both the robot and the surgical instrument used during the procedure. This allows the probe to repeat all movements made by the interventionist, enabling the acquisition of a US image centred on the surgical in instrument's point of action. The proposed system may facilitate the navigation of cardiovascular interventions using a transesophageal US probe.
13408-68
17 February 2025 • 5:30 PM - 7:00 PM PST | Golden State Ballroom
Show Abstract +
In this work, we propose and evaluate a benchtop testbed for robotic manipulation of an optical imaging probe. We use low-cost hardware and open-source software to construct the testbed and describe the implementation so that it can be easily adapted to other research areas.
13408-69
17 February 2025 • 5:30 PM - 7:00 PM PST | Golden State Ballroom
Show Abstract +
Estimating a patient’s auditory neural health is very important for cochlear implant (CI) patient-specific programming .One common method used to derive the number of healthy auditory nerve fibers (ANF) is through electrically evoked compound action potential (ECAP). Comparison between clinically measured ECAP responses and simulation can provide insights into a patient’s neural health condition. This simulation is currently done through in-house software that calculates simulated nerve fibers’ responses to stimulation but is very time-consuming. In this work we attempt to generate ECAP response using ResNet18 trained on the in-house ECAP simulations of 6 patients, augmented with varying resistivity parameters and CI locations. Our goal is to reduce computation time, allowing for model creation and patient-specific programming recommendations during a single patient visit, while also eliminating common problems with outliers and artifacts during ECAP simulation. Our method generated ECAP responses with peaks that were within 10% of the traditional model output on average, unaffected by stimulus artifacts, and with an average inference time of 1.12 ±0.08 s.
13408-70
17 February 2025 • 5:30 PM - 7:00 PM PST | Golden State Ballroom
Show Abstract +
This study aimed to improve the precision and efficacy of surgical treatment for developmental dysplasia of the hip (DDH) through a novel, patient-specific planning tool. Utilizing gait kinematics simulated on 3D reconstructed hip CT images, we developed a Bayesian optimization framework to extract features of articular cartilage stress and hip joint center (HJC) translation. We analyzed nine DDH cases, demonstrating that our method reduces HJC translation and cartilage stress, aligning with healthy hip biomechanics. This approach tailors PAO parameters to individual patients, offering a computationally efficient, automated tool for enhancing surgical planning and outcomes in DDH treatment.
13408-71
17 February 2025 • 5:30 PM - 7:00 PM PST | Golden State Ballroom
Show Abstract +
Computational tools, such as digital twin modeling, enable patient-specific surgical planning for ablative therapies in hepatocellular carcinoma treatment. Adaptive strategies, like multi-antenna thermal ablation, improve MWA utility for large tumors. The biophysical interaction of multiple ablation probes creates unique, fused ablation zones distinct from individual activations. Finite element modeling addresses this unpredictability but is time-consuming This study introduces a framework for biophysical digital twin models to forecast ablation volumes, training two machine learning models: PointNet Vector Model (PVM) and Fixed Grid Model (FGM) for rapid predictions. Preliminary data show PVM exhibits larger, more variable percentage volume differences compared to FGM. Initially, PVM differences range from -1.5% to 3%; FGM ranges from -1% to 1%. Over time, FGM error reduces to less than ±1% by 14-15 minutes, whereas PVM maintains differences from -1% to 2%. Neither model meets the expectation of 0% volume difference at every time point, indicating a need for improved training data simulations or reconceptualizing current methods.
13408-72
17 February 2025 • 5:30 PM - 7:00 PM PST | Golden State Ballroom
Show Abstract +
Percutaneous renal interventions are crucial for treating various renal diseases, with success highly dependent on the operator's skills. Ultrasound (US) imaging is widely used for diagnosing and monitoring these diseases, yet accurately interpreting US images is challenging. US-compatible phantoms are valuable tools for medical training and technology validation. This paper introduces a phantom designed with a mimic-breathing mechanism suitable for both US and computed tomography (CT) imaging. Utilizing tissue-mimicking materials with acoustic properties similar to human renal tissue, the phantom features an anatomically accurate kidney shape and a balloon-based bi-level positive airway pressure (Bi-PAP) system to replicate respiratory motion. Validation experiments confirm the model simulates respiratory-induced motion artifacts in US imaging. Quantitative analysis shows a strong correlation between imaging results from the phantom and clinical datasets, affirming its research and training suitability. Future work will simulate different steps of percutaneous intervention.
13408-73
17 February 2025 • 5:30 PM - 7:00 PM PST | Golden State Ballroom
Show Abstract +
Accurate retinal imaging is crucial for preclinical research and the development of diagnostic and therapeutic approaches for eye diseases. However, existing retinal imaging models often lack the precision needed to replicate the complex optical environment of the eye, limiting their effectiveness in testing and calibration. Current approaches rely heavily on live animals, which pose ethical concerns and variability in results. In this work, we designed a retinal phantom that can be used to validate spatial and spectral properties. We developed a 3D-printed mouse phantom retina to simulate the conditions of the eye to mimic its optical qualities. We produced gelatin phantom with deoxygenated blood to simulate veins and arteries. Imaging was performed using a topical endoscopic fundus imaging (TEFI) hyperspectral camera. For data analysis, we estimated the oxygenation rate in the phantom by assuming that oxygenated and deoxygenated blood are the two primary chromophores. The results showed that the phantom had rich both spatial and spectral details. Oxygenation mapping shows that the phantom produced reliable oxygenation rates of the blood.
13408-74
17 February 2025 • 5:30 PM - 7:00 PM PST | Golden State Ballroom
Show Abstract +
The development of accurate medical imaging phantoms is vital for advancing pre-clinical technologies, where precise simulation of human tissues is crucial for testing imaging modalities like ultrasound, magnetic resonance imaging (MRI), and computed tomography (CT). Polyvinyl alcohol cryogels (PVA-C) is a material useful for mimicking the mechanical and acoustic properties of human tissues after rounds of freeze-thaw cycles (FTC). This paper offers methodology considerations for preparing 10% w/w PVA-C phantoms optimized for these imaging parameters, focusing on PVA concentration, FTC, and the use of co-solvents and nanoparticles. By comparing techniques for liver and artery phantoms, the paper aims to achieve uniformity in phantom properties. Additionally, it explores PVA-C's self-healing and shape memory properties, relevant for biomedical implants and image-guided interventions. The full version of this paper will include specific details of best practices for manufacturing various organ phantoms.
13408-75
17 February 2025 • 5:30 PM - 7:00 PM PST | Golden State Ballroom
Show Abstract +
Brachytherapy involves bringing a radioactive source near tumor tissue using implanted needles. Image-guided brachytherapy planning requires among other, the reconstruction of the needles. Manually annotating these needles on patient images is a challenging and time-consuming task for clinicians. For automatic needle reconstruction, a two-stage pipeline is commonly adopted, comprising a segmentation stage followed by a post-processing stage. While deep learning models are effective for segmentation, they often produce errors such as small false-positively identified needle parts, under-segmentation, over-segmentation, and disconnected parts of needles. No currently existing post-processing technique is robust to all these possible segmentation errors. We develop an approach that can be used to improve the detection accuracy of existing post-processing techniques. Experiments on a prostate cancer dataset, based on MRI scans annotated by clinicians, demonstrate that our approach effectively manages segmentation errors.
13408-76
17 February 2025 • 5:30 PM - 7:00 PM PST | Golden State Ballroom
Show Abstract +
The segmentation of subcortical structures is critical in deep brain stimulation planning, enabling clinicians to visualise target regions and nearby important functional regions to avoid. However, due to the low contrast and small size of these regions, automatic segmentation on standard T1-weighted images often proves challenging. Recent interactive frameworks allow clinicians to correct automated segmentations using various interaction mechanisms, but there is little investigation into their usability, particularly with expert clinical users and mixed interaction methods. This article presents an observational study of three expert clinical users for various segmentation and segmentation correction tasks, measuring their objective level of use of different automatic and manual interaction mechanisms as well as their opinions on the segmentation interface. We have found that expert users appear to modify their preference for which interaction mechanisms based on the particular task at hand (i.e. segmentation from scratch or correcting an existing but erroneous segmentation) as well as their subjective expectations of the task.
13408-77
17 February 2025 • 5:30 PM - 7:00 PM PST | Golden State Ballroom
Show Abstract +
We propose a novel multi-stage ensemble optimization strategy to overcome the issue of local minima in 2D-to-3D registration of articulated joints. Inspired by block coordinate descent, our method uses multiple parallel optimizers on subsets of pose parameters, increasing the likelihood of finding the global optimum. When applied to image-guided wrist surgery, our technique achieved registration accuracy improvements of 13.59% to 33.95% over competing methods. This strategy shows promise for broader applications in surgical 2D-3D alignment problems.
13408-78
17 February 2025 • 5:30 PM - 7:00 PM PST | Golden State Ballroom
Show Abstract +
we propose a deep learning framework called the SAM-mixed network,which aims to enhance the accuracy and efficiency of semantic segmentation for polyps.
13408-79
17 February 2025 • 5:30 PM - 7:00 PM PST | Golden State Ballroom
Show Abstract +
This study developed two hybrid-based deep learning models, Hybrid-ConvKAN and Hybrid-MLP, for accurately segmenting cervical cancer tumors and organs from planning CT images. By integrating Kolmogorov-Arnold Convolutions and multilayer perceptrons, the models trained data from 157 patients, delineating 11 organs and clinical target volumes (CTVp and CTVn). The Hybrid-ConvKAN model achieved more accurate results with an average Dice Similarity Coefficient (DSC) of 0.91 for organs, 0.83 for CTVp, and 0.81 for CTVn, compared with individual models and Hybrid-MLP. These results suggest that Hybrid-ConvKAN model is promising for implementing dose stratification during the treatment.
13408-80
17 February 2025 • 5:30 PM - 7:00 PM PST | Golden State Ballroom
Show Abstract +
Cochlear implants (CIs) are neural prosthetics for patients with severe-to-profound hearing loss, consisting of an electrode array surgically implanted in the cochlea. The position of the array affects hearing outcomes, but traditional surgical approaches lack direct visualization, often leading to suboptimal placements of arrays. Our lab has proposed an augmented reality (AR) system to improve CI insertion using pre-operative CT-based planning and 3D to 2D registration with monocular microscope video for AR overlay. Herein, we hypothesize that the epitympanum is an ideal structure for CT and microscopy localization to aid in this registration. We propose and evaluate an atlas-based method of localizing the epitympanum in 3D CT images as the first step of this registration pipeline. This method builds upon our previously proposed DABS-LS framework, a 3D self-supervised U-Net-based network for accurate atlas-based segmentation. The method achieved median 95th percentile Hausdorff distances of only 1.05 mm compared to 78 test cases with manually labelled ground truth. These results demonstrate the promise of the method for use in augmented reality-assisted cochlear implantation.
13408-81
17 February 2025 • 5:30 PM - 7:00 PM PST | Golden State Ballroom
Show Abstract +
Pelvic magnetic resonance imaging (MRI) is a standard noninvasive imaging assessment of perianal fistulizing
Crohn’s Disease (CD-PAF). Perianal fistula classification, extent, volume, and response to therapy are difficult
to assess. Towards overcoming these issues, we aim to extensively evaluate standard deep-learning approaches
against more recent generalizable foundation models for the segmentation of fistulae and relevant anal canal
anatomy in an adult cohort of MRIs from 92 patients with CD-PAF. This could assist clinicians in guiding
management, evaluating fistula response to surgical therapy, and predictive image modelling. Both nnU-Net
and MedSam were able to robustly segment the internal sphincter muscle attaining a DSC of 0.92 ± 0.14 and
0.87 ± 0.13 (respectively), but the segmentations for the perianal fistula (0.42 ± 0.2 & 0.55 ± 0.09) and external
sphincter muscle (0.41 ± 0.16 & 0.12 ± 0.08) are sub-optimal compared to manual annotations. By evaluating the
performance of these segmentation models on CD-PAF, there is significant potential to improve the visualization
of the rectal structures.
13408-82
17 February 2025 • 5:30 PM - 7:00 PM PST | Golden State Ballroom
Show Abstract +
Federated Learning (FL) enables medical centers to collaboratively train deep-learning models while safeguarding patient data privacy. To access the efficiency of FL on 3D dose prediction, a key step in knowledge-based planning (KBP) for radiotherapy, we developed the FedKBP framework to evaluate the performances of centralized, federated, and individual (i.e. separated) training approaches on 340 treatment plans from OpenKBP dataset. The results showcase FL as a promising alternative to traditional pooled-data training, while posing the need of more sophisticated FL algorithms to handle inter-site data variation.
13408-83
17 February 2025 • 5:30 PM - 7:00 PM PST | Golden State Ballroom
Show Abstract +
Treatment of cervical cancer commonly involves high-dose-rate brachytherapy (HDR-BT), a procedure that requires precise and efficient planning to achieve the best patient outcomes. Historically, the HDR-BT planning process has been labor-intensive and largely dependent on the expertise of the clinician, resulting in potential inconsistencies in the quality of treatment. To overcome this issue, we propose an innovative method that employs advanced deep-learning models to improve HDR-BT planning. This paper presents the \textbf{Dendrite Cross-Attention UNet (DCA-UNet)}, which features a sophisticated dendritic structure comprising a primary branch for stacked inputs and three auxiliary branches dedicated to the segmentation of the clinical target volume (CTV), bladder, and rectum. This architecture enhances the model's understanding of organ-at-risk (OAR) areas, thereby improving dose prediction accuracy. Extensive evaluations reveal that DCA-UNet significantly enhances the precision of HDR-BT dose predictions across different applicator types. Our findings indicate that DCA-UNet consistently outperforms both traditional UNet and the more recent SwimUNetr models.
13408-84
17 February 2025 • 5:30 PM - 7:00 PM PST | Golden State Ballroom
Show Abstract +
The optimization of volumetric modulated arc therapy (VMAT) machine parameters is a time-consuming and challenging task due to the high dimensionality of the problem. Ideally, treatment plans should be adapted continually throughout the treatment to accommodate anatomical changes in the tumor and surrounding normal tissue. However, current workflow rarely include re-planning as it remains too time-consuming in practice. In this work, we propose a deep deterministic policy gradient framework to automatically generate deliverable VMAT plans using a multi-task CNN (MT-CNN) as policy network. This model consists of a single encoder coupled with two distinct decoder paths, enabling separate estimation of actions related to multi-leaf collimator positions and dose rate given as input a patient’s planning target volume (PTV) and organs at risk (OARs) contours. Preliminary results showed the proposed MT-CNN generated VMAT plans with an average run time of 3.11+/-0.76 seconds over our test set.
13408-85
17 February 2025 • 5:30 PM - 7:00 PM PST | Golden State Ballroom
Show Abstract +
Accurate positioning of radioactive sources in brachytherapy is crucial for effective tumor treatment while sparing normal tissue. This study presents a deep-learning-based method using a Generative Adversarial Network (GAN) to enhance limited-angle cone-beam CT (CBCT) images, trained on 66 non-contrast pelvic CT scans. The method significantly improved the similarity index from 0.50 (Filtered Back Projection) to 0.85, outperforming traditional techniques in accuracy and speed. A prototype C-arm CT/SPECT system with a DRTECH detector successfully distinguished geometric structures and materials of similar densities. Real-time Co-60 source tracking was achieved using SPECT imaging, enhanced by an in-house algorithm. These advancements demonstrate the potential for integrating this technology into clinical settings, improving brachytherapy's precision and efficacy with real-time dose monitoring and internal condition verification.
13408-86
17 February 2025 • 5:30 PM - 7:00 PM PST | Golden State Ballroom
Show Abstract +
Real-time tumor tracking can monitor patient motion, verify tumor position, and guide the radiation beam to the tumor target in X-ray-guided radiation therapy. However, markerless kilovoltage (kV) X-ray image-based tumor tracking is a longstanding challenge in clinical practice due to its low tumor contrast and visibility. In this work, we aim to develop a patientspecific conditional diffusion model to generate the decomposed tumor image (DTI) for X-ray-guided tumor tracking in radiotherapy. The proposed method is evaluated using lung patients and can effectively improve the target contrast of lung tumors on the kV projection images.
13408-87
17 February 2025 • 5:30 PM - 7:00 PM PST | Golden State Ballroom
Show Abstract +
Lung injuries such as radiation induced lung injuries (RILI) can lead to heterogeneous alterations to lung biomechanics. Such injuries can affect lung function, leading to a lack of oxygenation, which might require mechanical ventilation. However, improper mechanical ventilation can lead to further injuries to the lungs, termed ventilator induced lung injuries. We hypothesize that such injuries can be quantified as changes in the motion of lung tissue during respiration. In this study, we present an investigation into the evolution of kinematic biomarkers in a rat model of RILI prior to and post-radiation. We performed dynamic imaging on a healthy Sprague-Dawley rat before and after irradiation. We quantified the alteration of kinematic biomarkers and observed increased volumetric strain and heterogeneity in the biomarkers. We expect such kinematic biomarkers to help improve our understanding of lung injuries, their evolution, and their effect on lung health.
13408-88
17 February 2025 • 5:30 PM - 7:00 PM PST | Golden State Ballroom
Show Abstract +
In this study, we introduce a set of patient-specific lung diffusion models (PS-LDMs) that utilize diffusion models combined with LoRA fine-tuning to generate sCT images from CBCT images. The enhanced HU accuracy of the sCT images compared to pre-correction CBCT demonstrates a significant improvement in image quality, with sCTs approaching the quality of dpCT images. By leveraging prior CBCT fractions to correct artifacts in current treatment CBCT images and capture inter-fraction anatomical changes, we can create a tailored model for each patient undergoing RT. We evaluated the PFS-LDMs using an institutional lung cancer dataset, assessing their performance with metrics as mean absolute error (MAE), peak signal-to-noise ratio (PSNR), normalized cross-correlation (NCC), and structural similarity index measure (SSIM).
13408-89
17 February 2025 • 5:30 PM - 7:00 PM PST | Golden State Ballroom
Show Abstract +
In this study, we investigated a novel hybrid foundation model (HFM) by combing foundation anatomy model and foundation segmentation model (FAM-FSM) for automatic object segmentation on thoracic CT images. The previous hybrid intelligence (HI) segmentation system using natural intelligence through anatomy models to provide recognition results as prompts to FSM enables combination of recognition with delineation from FSM for object segmentation. SAM-Med2D is adopted as FSM for segmentation. HFM achieved promising results compared to FSM using manual ROI prompts. Fine-tuning FSM might not perform well on challenging objects compared to small but specific trained models.
13408-90
17 February 2025 • 5:30 PM - 7:00 PM PST | Golden State Ballroom
Show Abstract +
Myocardial infarction (MI) leads to scar formation that significantly alters cardiac mechanics. This study examines the effects of infarct area size and fiber alignment on heart function. Results show that larger infarct areas and increased fiber disarray reduce healthy left ventricle (LV) contractile behavior and increase fiber strain in the infarcted region. Misaligned and larger infarct areas deviate both infarcted and healthy myocardial zones from their expected functions, underscoring the importance of scar geometry and fiber alignment in post-MI cardiac performance.
13408-91
17 February 2025 • 5:30 PM - 7:00 PM PST | Golden State Ballroom
Show Abstract +
Coronary artery bypass grafting (CABG) is a form of cardiac surgery traditionally performed with the heart stopped. The off-pump CABG (OP-CABG) technique, which keeps the heart beating, offers superior long-term results and is increasingly preferred. Effective training for OP-CABG is crucial, but existing tools fall short in accurately simulating a beating heart. Thus, we propose a mechanical training device with disposable synthetic coronary arteries on a silicone heart surface. Our device simulates cardiac movement through the timed tensioning and releasing of pull cords, driven by a servo-motor. An optical tracking system demonstrates that our model effectively replicates the step displacements and range of motion reported in literature, making it a valuable training tool for the OP-CABG procedure.
13408-92
17 February 2025 • 5:30 PM - 7:00 PM PST | Golden State Ballroom
Show Abstract +
Proton beam irradiation to the ventricular myocardium is a promising non-invasive treatment for ventricular tachycardia. Prior work has demonstrated that lesions can be visualized in myocardial tissue using delayed contrast enhanced magnetic resonance imaging (DCE-MRI) following proton beam therapy; however, an automatic method for measuring these lesions has not yet been described. Enhanced regions in DCE-MRI can be defined by delineating a remote region (distant from scar or enhanced region) and then thresholding at 1-6 standard deviations (SD) above the mean voxel intensity in the remote myocardium. The current study aims to evaluate the appropriate threshold for automatic measurement of lesions from DCE-MRI in ventricular myocardium following proton beam therapy.
13408-93
17 February 2025 • 5:30 PM - 7:00 PM PST | Golden State Ballroom
Show Abstract +
Occluding the LAA with percutaneous devices is now a standard clinical practice for patients with non-valvular atrial fibrillation to reduce the risk of stroke. These devices have pre-defined models in terms of size and require therefore, during the intervention, extraction of specific measurements of the LAA. Manual device sizing is the standard practice, however semi-automatic methods have also been explored for 3D medical images.
Recently, the authors have researched the potential of deep learning methods to segment the LAA in 2D TEE. In this study, a pipeline to extract the clinical measurements is described. A total of 4 different input methodologies were studied, 3 of them using independent anatomical feature maps and a last one combining all described feature maps.
Three metrics were used to evaluate the performance of each methodology: the first was the angulation between the Ground Truth and the Predict, the second was the distance between the midpoint of each line, and the last was the different in terms of clinical indexes, namely landing zone. The results showed high agreement between automatic and manual LAA clinical indexes for the landing zone diameter.
13408-94
17 February 2025 • 5:30 PM - 7:00 PM PST | Golden State Ballroom
Show Abstract +
This study introduces a reinforcement learning model designed to enhance the precision and safety of catheterization in percutaneous coronary interventions (PCI). Utilizing Unity 3D and ML-Agents, the model significantly reduces procedural risks and costs while improving efficiency and quality. Experimental results indicate its adaptability and potential for optimization in various clinical scenarios, marking a significant advancement in AI integration into cardiac healthcare.
13408-95
17 February 2025 • 5:30 PM - 7:00 PM PST | Golden State Ballroom
Show Abstract +
This study highlights the significant impact of thoracic skeletal abnormalities on respiratory function in pediatric patients with Thoracic Insufficiency Syndrome (TIS). Utilizing free-breathing dynamic MRI (dMRI), we non-invasively assessed and quantified respiratory volumes, revealing that VEPTR surgery markedly improves volumes, especially right hemi-diaphragm tidal volume. The correlation between thoracic kyphosis and right-sided respiratory volumes underscores the critical role of spinal curvature in respiratory outcomes. The lack of significant correlations between lumbar lordosis and respiratory volumes suggests thoracic spinal deformities have a more pronounced effect on lung function.
13408-96
17 February 2025 • 5:30 PM - 7:00 PM PST | Golden State Ballroom
Show Abstract +
Ponseti Casting is the gold standard for treating clubfoot deformity in kids, in which successive plaster casts are placed to bring the deformed foot to its natural position. The deformity evaluation is based on certain angles that physicians use to plan the treatment. Severity is often classified by the DiMeglio and Pirani scoring systems. However, there is debate about their subjective nature and effectiveness, as these scores primarily provide an approximate prediction of the number of casts needed for deformity correction rather than a progressive correction plan. Since imaging modalities are usually not involved in clubfoot correction, no data is available to quantify the deformity correctly and to learn about the casting process. Therefore, we have developed an accurate method to quantify the clubfoot deformity using the 3D-reconstructed images of the patient’s scans. We have also developed a MATLAB application that uses the 3D models to calculate the clubfoot angles.
13408-97
17 February 2025 • 5:30 PM - 7:00 PM PST | Golden State Ballroom
Show Abstract +
An alveolar molding plate treatment (AMP) can near-normalize the oral cleft deformity of neonates with unilateral cleft lip and palate (UCLP) prior to their first cleft repair surgery (cleft lip repair). However, its ability to produce complete closure of the alveolar cleft based on pre-treatment conditions is unknown. We aimed to test if a machine learning algorithm can be used to help predict the outcomes of an AMP called Biocreative AMP (BioAMP). The model was trained on 16 landmarks and 137 parameters. The goal was to classify the patients' treatment outcomes based on the observed morphological changes and birth weight data. The model reported an accuracy of 81% in predicting treatment outcomes for neonates with complete UCLP. The highest importance was observed for birth weight, then the distance between fiducial point pairs across the alveolar gap. Outlier weights were more likely to have bad outcomes than the weights closer to the mean. The results demonstrate that the machine learning pipeline we developed may be a viable tool to help inform treating clinicians about necessary pre-treatment modifications on BioAMP to produce complete closure of the alveolar cleft.
13408-98
17 February 2025 • 5:30 PM - 7:00 PM PST | Golden State Ballroom
Show Abstract +
We evaluated effects of pre-operative QdMRI on surgery planning in TIS patients, where patients were categorized into two groups: Change group (Gc) where surgery planning would have changed based on knowledge of QdMRI parameters, and no change group (Gnc) otherwise. Post-operative dMRI was used to evaluate surgery effects. Age-adjusted Mahalanobis distances for right lung height and left lung volumes were significantly and borderline non-significantly larger, respectively, in Gc, indicating greater deviations from normality in those where surgical planning would have changed based on knowledge of pre-operative QdMRI.
Tuesday Morning Keynotes
18 February 2025 • 8:30 AM - 10:00 AM PST | Town & Country B/C
Session Chairs:
Jhimli Mitra, GE Research (United States), Christian Boehm, ETH Zurich (Switzerland)
View Full Details: spie.org/tuesday-morning-keynotes
8:30 AM - 8:35 AM:
Welcome and introduction
8:35 AM - 8:40 AM:
Award announcements
- Robert F. Wagner Award finalists for conferences 13406 and 13412
- Image Processing Student Paper Award
13406-504
Integrating computational approaches to unravel alzheimer's disease and co-pathologies: a biomarker-driven approach to precision medicine
(Keynote Presentation)
18 February 2025 • 8:40 AM - 9:20 AM PST | Town & Country B/C
Show Abstract +
Early detection and intervention in neurodegenerative diseases hold the potential to significantly impact patient outcomes. This presentation will explore the development of multi-disciplinary and multi-modality biomarkers to identify individuals at risk and monitor disease progression. By combining advanced imaging techniques, such as MRI, and PET, with fluid biomarkers, we aim to detect subtle changes in brain structure and function that precede clinical symptoms. These biomarkers could serve as powerful tools for early diagnosis, enabling timely intervention and potentially delaying disease onset. Furthermore, by identifying individuals at highest risk, we can optimize the design of clinical trials and accelerate the development of effective therapies. Ultimately, our goal is to improve the lives of individuals with neurodegenerative diseases through early detection, precise diagnosis, and targeted treatment.
13412-505
Wearable ultrasound technology
(Keynote Presentation)
18 February 2025 • 9:20 AM - 10:00 AM PST | Town & Country B/C
Show Abstract +
The use of wearable electronic devices that can acquire vital signs from the human body noninvasively and continuously is a significant trend for healthcare. The combination of materials design and advanced microfabrication techniques enables the integration of various components and devices onto a wearable platform, resulting in functional systems with minimal limitations on the human body. Physiological signals from deep tissues are particularly valuable as they have a stronger and faster correlation with the internal events within the body compared to signals obtained from the surface of the skin. In this presentation, I will demonstrate a soft ultrasonic technology that can noninvasively and continuously acquire dynamic information about deep tissues and central organs. I will also showcase examples of this technology's use in recording blood pressure and flow waveforms in central vessels, monitoring cardiac chamber activities, and measuring core body temperatures. The soft ultrasonic technology presented represents a platform with vast potential for applications in consumer electronics, defense medicine, and clinical practices.
Break
Coffee Break 10:00 AM - 10:30 AM
Session 4:
Cardiac Applications
18 February 2025 • 10:30 AM - 12:10 PM PST | Town & Country D
Session Chairs:
Cristian A. Linte, Rochester Institute of Technology (United States), David R. Holmes, Mayo Clinic (United States)
13408-16
18 February 2025 • 10:30 AM - 10:50 AM PST | Town & Country D
Show Abstract +
Minimally invasive surgery is the state-of-the-art approach for mitral valve repair, but altered pressure conditions during surgery deform the anatomy, complicating visualization and measurement. We developed a technique combining stereo-endoscopic video with 3D transesophageal echocardiography (3D TEE) to enhance anatomic visualization and measurement accuracy. Our method includes stereo camera calibration, image segmentation, and 3D model reconstruction, aligning imaging modalities with anatomical landmarks. Validation showed high precision within an error range of 0.5 ± 0.1 mm. Integrating stereoscopic and 3D TEE offers improved precision in mitral valve repairs, with potential future applications in visualizing tissue properties and implant placement.
13408-17
18 February 2025 • 10:50 AM - 11:10 AM PST | Town & Country D
Show Abstract +
With the popularity of Competency-based Medical Education (CBME) and Artificial Intelligence (AI) rising, there is an increased need for large robust training datasets for surgical simulators. This need is emphasized by
the current scarcity of publicly available data featuring real surgical human Mitral Valves (MVs) and ‘dry lab’
surgical MV phantoms to generate realistic frames for MV simulators. We present a comprehensive large dataset consisting of 3 parts: extracted frames from MV surgical videos, frames from MV Phantom videos, and images generated using unpaired GAN models. By using the proposed dataset to generate realistic MV image frames during surgical training, surgeons can accurately practice MV repair surgeries with a more realistic, high resolution and stable framed visual appearance of the MV. Consequently, users immersed in this realistic training experience can potentially be better prepared for Minimally Invasive MV surgeries.
13408-18
18 February 2025 • 11:10 AM - 11:30 AM PST | Town & Country D
Show Abstract +
Ventricular septal defect surgery requires three-dimensional visualization skills from surgeons during a hybrid procedure. The Apple Vision Pro has the potential to assist surgeons in such surgeries because of its reliance on eye-tracking technology. This research aims to provide a quantitative assessment of the Vision Pro’s eye-tracking capacities to assess whether it can be a good tool to use in a medical context. Our pilot study showed that the Vision Pro’s accuracy was better than most other VR/AR headsets when collecting data from a random saccades task. We intend on running this study with more participants as well as extending it to a more real-world clinical setting.
13408-19
18 February 2025 • 11:30 AM - 11:50 AM PST | Town & Country D
Show Abstract +
Accurate left atrium (LA) segmentation from pre-operative scans is crucial for diagnosing atrial fibrillation, treatment planning, and supporting surgical interventions. While deep learning models are key in medical image segmentation, they often require extensive manually annotated data. Foundation models trained on larger datasets have reduced this dependency, enhancing generalizability and robustness through transfer learning. We explore DINOv2, a self-supervised learning vision transformer trained on natural images, for LA segmentation using MRI. The challenges for LA’s complex anatomy, thin boundaries, and limited annotated data make accurate segmentation difficult before & during the image-guided intervention. We demonstrate DINOv2’s ability to provide accurate & consistent segmentation, achieving a mean Dice score of .871 & a Jaccard Index of .792 for end-to-end fine-tuning. Through few-shot learning across various data sizes & patient counts, DINOv2 consistently outperforms baseline models. These results suggest DINOv2 effectively adapts to MRI with limited data, highlighting its potential as a competitive tool for segmentation & encouraging broader use in medical imaging.
13408-20
18 February 2025 • 11:50 AM - 12:10 PM PST | Town & Country D
Show Abstract +
Cardiovascular disease (CVD) causes approximately 50% of deaths in cases of chronic kidney disease (CKD). Myocardial fibrosis is a common occurrence in individuals with CKD and is a strong predictor of adverse cardiovascular events/death. Standard of care cardiac imaging is limited by inter-reader variability, so we propose a novel proteomics-driven radiomic signature for myocardial fibrosis in end-stage kidney disease patients, via CMR T1 maps. Through the use of a dataset of CKD patients who were imaged at diagnosis and after 9 months, we identified imaging markers associated with proteomic MCSF (macrophage colony stimulating factor) and kidney transplant within the myocardial wall. When evaluated via a random forest classifier, we achieved an AUC of 0.75 and 0.78 in cross-validated training using baseline and follow-up scans respectively with validation AUCs of 0.17 and 0.83. Our novel multimodal feature interrogation approach may enable improved characterization of myocardial fibrosis in end-stage patients.
Break
Lunch Break 12:10 PM - 1:40 PM
Session 5:
Image-Guided Procedures, Robotic Interventions, and Ultrasonic Imaging/Tomography: Joint Session with Conferences 13408 and 13412
18 February 2025 • 1:40 PM - 3:00 PM PST | Town & Country D
Session Chair:
Jessica R. Rodgers, Univ. of Manitoba (Canada)
13412-7
18 February 2025 • 1:40 PM - 2:00 PM PST | Town & Country D
Show Abstract +
In this presentation, we present a modular ultrasound (MODUS) framework, allowing clinicians to have an ultrasound array in arbitrary aperture size and on flexible surface curvatures, which can maintain effectiveness regardless of patient age and physical condition.
13408-21
18 February 2025 • 2:00 PM - 2:20 PM PST | Town & Country D
Show Abstract +
Real-time intraoperative localization during breast cancer surgery is essential to ensure complete tumor resection. Unfortunately, limitations in existing single-modality devices result in high rates of revision surgeries. This research details the design and fabrication of a novel hybrid imaging system that mechanically couples a unique type of focused gamma probe with an ultrasound transducer to simultaneously acquire anatomical and functional images in real time. Compared to existing literature, phantom studies of our radio-ultrasound guided system revealed a substantial improvement in resolution (almost an order of magnitude better) while maintaining high sensitivity. The complementary information provided by precisely visualizing ‘hot’ radiolabeled tissues within the anatomy is expected to improve the accuracy of breast cancer surgery, leading to improved patient outcomes.
13412-8
18 February 2025 • 2:20 PM - 2:40 PM PST | Town & Country D
Show Abstract +
Midline shift (MLS) is a diagnostic marker for brain pathologies such as intracranial hemorrhage and traumatic brain injury, requiring emergent treatment. CT and MRI are the gold standards for MLS measurement but prevent repeated, short-interval monitoring and are associated with significant risks when transporting critically ill patients. This feasibility study investigates a freehand 3D ultrasound (US) system for bedside MLS measurement. Using electromagnetic position trackers on both the US transducer and the subject, we reconstructed 2D B-mode ultrasound images into a 3D volume. We scanned a healthy subject through bilateral transtemporal windows and measured MLS on the reconstructed volume. Our system achieved a measurement error comparable to existing sonography-based methods, fitting clinical requirements, and proving its feasibility for clinical use. Our approach reduces operator-skill dependency and provides a rapid, non-ionizing solution for continuous, short-interval monitoring of MLS, potentially enhancing clinical decision-making in critical care and emergency field diagnosis.
13408-22
18 February 2025 • 2:40 PM - 3:00 PM PST | Town & Country D
Show Abstract +
Smaller, affordable ultrasound equipment is widely used, but image quality issues require specialized training. Augmented reality (AR) enhances training but faces challenges in development tools, real-time tracking, and costs. The proposed system utilizes Meta Quest3, interacting with the Clarius L15, an ultrasound simulator. The AR platform employs Touch Plus controllers to track the US probe and biopsy needle, allowing the physician to maneuver the probe freely regardless of movement. To evaluate this method's viability, the accuracy and precision of the Touch Plus controllers in instrument tracking were studied and compared to a commercial electromagnetic tracking system. Experiments showed that the proposed strategy achieved an accuracy of 0.6858°±0.5531° and a precision of 0.3640°±0.030° in 12 different orientations and positions. The results demonstrated the proposed approach's effectiveness in tracking instruments without external markers in various orientations and positions, indicating its potential for clinical practice.
Break
Coffee Break 3:00 PM - 3:40 PM
Session 6:
Physics/Image-Guided Procedures: Joint Session with Conferences 13405 and 13408
18 February 2025 • 3:40 PM - 5:20 PM PST | Town & Country B
Session Chairs:
Shuai Leng, Mayo Clinic (United States), Shuo Li, Case Western Reserve Univ. (United States)
13405-26
18 February 2025 • 3:40 PM - 4:00 PM PST | Town & Country B
Show Abstract +
We introduce a novel learning-based approach for rigid patient motion estimation in interventional C-arm CBCT. Several motion compensation strategies exist for different domains of the imaging process, including reconstruction-free data consistency assessments, using epipolar consistency conditions, and reconstruction-based autofocus methods, optimizing image quality metrics. Consistency-based approaches are merely sensitive to out-of-plane motion, while autofocus methods can detect in-plane motion but tend to converge into local minima for larger motion amplitudes. We aim to leverage the strengths of both methods by combining them in a deep learning-based model. Our model consists of two parallel feature extraction networks followed by a classifier, and is trained and tested on simulated motion trajectories applied to motion-free clinical acquisitions. The network predicts 6 rigid directional displacement probabilities for each of the 496 projections. First tests achieve a good classification performance with an average ROC-AUC of 0.9527 and PR-AUC of 0.3961.
13408-23
18 February 2025 • 4:00 PM - 4:20 PM PST | Town & Country B
Show Abstract +
Transcatheter arterial embolization procedures have become more complex and widely used, but the lack of quantitative measures for embolization endpoints limits standardization, relying heavily on the experience of interventional radiologists. In this work we use a benchtop flow system and a 3D printed phantom of hepatic arterial vasculature to build an in vitro model of the dynamic conditions of the blood flow proximal to the catheter tip location during transcatheter embolization of the hepatic arteries. By controlling flow rates and using optical imaging, reproducible vessel occlusion conditions are simulated. The derived image-based metrics, such as time constants and area under the curve of the time-dependent image intensity profile, show consistent patterns and a strong correlation with occlusion levels, suggesting their potential as quantitative measures for embolization endpoints.
13405-27
18 February 2025 • 4:20 PM - 4:40 PM PST | Town & Country B
Show Abstract +
Quantitative digital subtraction angiography (qDSA) is a method for measuring blood velocity from 2D x-ray sequences in interventional radiology. Previous studies in swine models and phantoms compared qDSA to MRI or ultrasound. This study reports a computational fluid dynamics (CFD) simulation platform for controlled investigations of qDSA behavior. Iodine injections into arterial blood flow were simulated with OpenFOAM for different catheter geometries (0-45° angle, 0-180° rotation) and blood velocities. qDSA was applied to projections derived from CFD results. qDSA was linearly related to downstream CFD velocity and catheter orientation changes resulted in 5% standard deviation in qDSA velocity estimates.
13408-24
18 February 2025 • 4:40 PM - 5:00 PM PST | Town & Country B
Show Abstract +
In this work, we developed a deep CNN-based spectral metal artifact reduction algorithm for interventional oncology applications (sMARIO) and quantitatively assessed its performance, with a focus on iodine detection. Patient data was used to train the algorithm, which was subsequently applied to phantom images containing water and iodinated rods adjacent to metallic ablation applicators and used to assess its performance. Spectral MARIO effectively corrected substantial contributions of metal artifacts in the image data across all keVs investigated (40-150 keV). The resulting performance of sMARIO on 70 keV images and below (AUC=0.98) demonstrates the ability to accurately classify small iodine concentrations even in close proximity of metallic objects used in IO procedures.
13405-28
18 February 2025 • 5:00 PM - 5:20 PM PST | Town & Country B
Show Abstract +
We have developed an algorithm called IPEN (intra-intervention perfusion assessment using a standard x-ray system with no gantry rotation) to allow for perfusion assessment in interventional (IR) suites. IPEN puts this 4-dimensional (4-D = 3-D plus time) image reconstruction problem into a well-posed problem framework such that there exists a robust and rigorous solution. We have validated the accuracy with brain and liver perfusion applications. Recently we have further developed IPEN to address patient motion problems, as motion artifacts are common in clinical settings. The method called MC–IPEN uses native (i.e., non-subtracted) x-ray angiography (XA) images and compensates for the effect of head motion very effectively. The use of XA images, however, could be a problem for many hospitals, because they save digital subtraction angiography (DSA) images routinely but not XA images. In this work, we will further develop MC–IPEN such that it works with DSA images and does not need XA images.
NIH/NIBIB Session: Funding Opportunities and Grant Writing Tips for New Investigators
18 February 2025 • 5:30 PM - 6:30 PM PST
Session Chairs:
John M. Sabol, Konica Minolta Healthcare Americas, Inc. (United States), Maryam E. Rettmann, Mayo Clinic (United States)
View Full Details: spie.org/nih-nibb-session
5:30 PM – 6:00 PM
Navigating the NIH Grant System and Tips for Preparing Successful and Competitive NIH Grant Applications
Speaker: Behrouz Shabestari, Director, NIBIB National Technology Centers Program; Director, Division of Health Informatics Technologies (DHIT), NIBIB
6:00 PM – 6:15 PM
Scientific Program and Funding Opportunities at NIBIB
Speaker: Rui Pereira de Sá, Program Director, Division of Health Informatics Technologies (DHIT), NIBIB
6:15 PM – 6:30 PM
Question and Answers with Behrouz Shabestari and Rui Pereira de Sá
Wednesday Morning Keynotes
19 February 2025 • 8:30 AM - 10:00 AM PST | Town & Country B/C
Session Chairs:
Maryam E. Rettmann, Mayo Clinic (United States), Aaron D. Ward, The Univ. of Western Ontario (Canada)
View Full Details: spie.org/wednesday-morning-keynotes
8:30 AM - 8:35 AM:
Welcome and introduction
8:35 AM - 8:40 AM:
Award announcements
- Robert F. Wagner Award finalists for conferences 13408 and 13413
- Early-Career Investigator Award: Image-Guided Procedures, Robotic Interventions, and Modeling
- Student Paper Award: Image-Guided Procedures, Robotic Interventions, and Modeling
13408-506
Image-guided surgery and examinations using ultrasound
(Keynote Presentation)
19 February 2025 • 8:40 AM - 9:20 AM PST | Town & Country B/C
Show Abstract +
Many of today’s cancer surgeries are carried out with robot assistance. Using real-time intra-operative ultrasound, we can overlay pre-operative imaging into the surgeon’s console, enabling visualization of sub-surface anatomy and cancer at the same time with the standard laparoscopic camera view. We will discuss aspects of system design, visualization and registration methods that enable such visualization, and present our results. We will also present tissue and instrument tracking approaches that can be used in future augmented reality systems.
For remote and underserved communities, we developed a teleultrasound approach that relies upon using a novice – the patient, a family member or friend – as a robot to carry out the examination. The novice wears a mixed reality headset and follows a rendered virtual ultrasound transducer with the actual transducer. The virtual transducer is controlled by an expert, who sees the remote ultrasound images and feels the transducer forces. This tightly-coupled expert-novice approach has advantages relative to both conventional and robotic teleultrasound. We discuss our system implementation and results.
13413-507
The future of diagnostics: the role of computational pathology in tomorrow's medicine
(Keynote Presentation)
19 February 2025 • 9:20 AM - 10:00 AM PST | Town & Country B/C
Show Abstract +
Computational Pathology has already led to remarkable innovations in diagnostics, achieving expert pathologist performance in tasks such as prostate cancer grading and cancer metastasis detection. In recent years, we have seen rapid advances, with weakly supervised models able to predict patient outcomes or genetic mutations and foundation models enabling application to rarer diseases. However, this only scratches the surface of what will be possible in the near future.
In this talk, I will briefly touch on the history of computational pathology and how we got to where we are today. Subsequently, I will highlight the current methodological innovations in the field and their potential for causing a paradigm shift in diagnostic pathology. I will discuss how these innovations, combined with the AI-driven integration of radiology, pathology, and 'omics data streams, could change the future of diagnostics as a whole. Last, I will discuss the challenges and pitfalls moving forward and how we, as a community, can contribute to addressing them.
Break
Coffee Break 10:00 AM - 10:30 AM
Session 7:
Surgical Data Science
19 February 2025 • 10:30 AM - 12:10 PM PST | Town & Country D
Session Chairs:
Pierre Jannin, Lab. Traitement du Signal et de l'Image (France), Jeffrey Harold Siewerdsen, The Univ. of Texas M.D. Anderson Cancer Ctr. (United States)
13408-25
19 February 2025 • 10:30 AM - 10:50 AM PST | Town & Country D
Show Abstract +
Accurate surgical phase recognition from surgical videos is critical for surgical training and evaluation. However, traditional models like Transformers face limitations in memory and computational efficiency, particularly as input sequence length increases. We introduce an enhanced Sequential Picking Strategy with hyperparameter optimization within the ESPT-SVNet model, which integrates spatial and temporal embeddings. Through hyperparameter search, we are able to apply different sequential picking strategies by optimizing these hyperparameters, which allows the model to selectively extract crucial features from temporal embeddings, thereby reducing computational and memory demands to varying degrees. By optimizing these hyperparameters, we achieved significant performance improvements, reducing memory usage by 8.6% and training time by 16.2%, while maintaining or even enhancing model accuracy. We evaluated our approach using the ESPT-SVNet model. Evaluations on the Cholec80 dataset demonstrated remarkable results, with our method achieving an accuracy of 90.5% and a precision of 92.4%, surpassing existing models.
13408-26
19 February 2025 • 10:50 AM - 11:10 AM PST | Town & Country D
Show Abstract +
Robotic-assisted minimally invasive esophagectomy (RAMIE) is an established treatment for esophageal cancer, offering superior patient outcomes compared to open surgery and traditional minimally invasive surgery. This procedure is highly complex, involving the manipulation of multiple anatomical areas and navigating through repetitive phases and non-sequential phase transitions. Our objective is to utilize deep learning for surgical phase recognition in RAMIE to provide intraoperative support to surgeons. To this end, we have developed a novel surgical phase recognition dataset specifically tailored to RAMIE. Using this dataset, we conducted a benchmark study of state-of-the-art surgical phase recognition models. To better capture the temporal dynamics of this complex procedure, we developed a new model featuring an encoder-decoder structure with causal hierarchical attention, which demonstrates superior performance in recognizing surgical phases.
13408-27
19 February 2025 • 11:10 AM - 11:30 AM PST | Town & Country D
Show Abstract +
A computational framework for statistical surgical process modeling (SPM) was used to quantitatively evaluate performance in transbronchial lung nodule biopsy under fluoroscopy + EBUS compared to 3D cone-beam CT (CBCT) guidance and robotic assistance. Clinical outcomes were measured and validated in terms of procedure cycle time, radiation dose, and diagnostic yield, demonstrating quantitative gains in diagnostic yield for 3D guidance and helping to identify optimal workflow of these emerging technologies.
13408-28
19 February 2025 • 11:30 AM - 11:50 AM PST | Town & Country D
Show Abstract +
An automated method for evaluating surgical skills using optical flow and self-attention mechanisms is presented. Unlike previous methods that rely on kinematic data from surgical robots, this approach uses only surgical video as input. Optical flow is leveraged to extract motion information, and the ViViT model captures features from both image and motion data. These features are combined to predict six scores based on the OSATS (Objective Structured Assessment of Technical Skills) criteria. Evaluated on the JIGSAWS dataset, the model achieves a low root mean square error (RMSE), demonstrating its effectiveness in generating reliable skill assessments and its potential for broader applications in surgical evaluation.
13408-29
19 February 2025 • 11:50 AM - 12:10 PM PST | Town & Country D
Show Abstract +
We propose using a GAN-based image-to-image translation method to enhance the realism of surgical training phantom images. Our goal is to make these phantoms more suitable as customized training tools and to increase their utility for various AI applications. The Cycle-GAN model is employed to translate images from the intra-operative domain to the phantom domain. The data used includes both real surgeries and phantom images from laparoscopic prostatectomy procedures. A secondary action detection task is then used to evaluate the quality of our generated realistic phantom images. Study outcomes suggest that generating realistic synthetic data is feasible and can potentially improve deep learning-based action detection in minimally invasive surgery, paving the way for advancements in intelligent surgical assistants.
Break
Lunch Break 12:10 PM - 1:40 PM
Session 8:
Robotic Interventions
19 February 2025 • 1:40 PM - 3:00 PM PST | Town & Country D
Session Chairs:
Robert J. Webster, Vanderbilt Univ. (United States), Ziv R. Yaniv, National Institute of Allergy and Infectious Diseases (United States)
13408-30
19 February 2025 • 1:40 PM - 2:00 PM PST | Town & Country D
Show Abstract +
Neurosurgeons often desire to deliver needles through nonlinear paths into the brain. And in the case of thermal
therapy, once the needle tip is positioned correctly, they desire conformable control of both the size and shape of
the treatment zone. In this paper, we propose a new laser-based probe design that can enter the brain through
a nonlinear trajectory. Once at the target, this probe can deliver energy perpendicular to its path. This enables
shaping of a thermal treatment zone of complex geometry through a combination of probe rotation and insertion,
between subsequent periods of delivering laser energy. The motivation for our work on this probe is a larger
project related to thermal therapy of the hippocampus to treat epilepsy, which we describe as an example.
13408-31
19 February 2025 • 2:00 PM - 2:20 PM PST | Town & Country D
Show Abstract +
40% of breast cancer patients require corrective surgery after tumour resection via breast-conserving surgery. Sweeping of the resection cavity by spectroscopy and ultrasound is emerging as a solution for identifying residual cancer. However, this is challenging as breast tissue moves frequently. We present an approach for tracking the motion of a resection cavity with a robotic arm. We use EM tracking and a 6-axis robotic arm to track a simulated resection cavity. We embed an EM sensor in a retractor that holds the cavity open. A 3D Slicer module is then used to detect motion from the sensor and command the arm to follow the movement. To assess this approach we move the retractor to 36 positions in the workspace and record the latency between when a command is published and when the robot moves. We attach a camera to the end-effector of the robot to determine when the robot has successfully tracked the cavity by checking if it is visible in the camera frame. The latency was recorded to be 832.1ms on average. We can also successfully track the motion of the cavity 92% of the time. These results suggest that tracking of the breast cavity using EM tracking and robotics is feasible.
13408-32
Comparing freehand and robotic techniques for standard scan plane localization in ultrasound imaging
19 February 2025 • 2:20 PM - 2:40 PM PST | Town & Country D
Show Abstract +
The choice between freehand and robotic ultrasound depends on the specific use case, the environment, and the priorities of the healthcare provider. This work compares and evaluates these two approaches for localizing scan planes during renal imaging. A prototype system, integrated with a reinforcement learning agent, was developed, capable of emulating inconsistencies in freehand imaging, such as deviations in executing motions and maintaining contact forces. Evaluations on phantom and cadaver specimens showed comparable performance and demonstrated the relative robustness of the freehand approach, suggesting its suitability for diagnostic and point-of-care settings where robotic solutions may not be practical.
13408-33
19 February 2025 • 2:40 PM - 3:00 PM PST | Town & Country D
Show Abstract +
Esophageal cancer is among the most common types of cancer worldwide. It is traditionally treated using open esophagectomy, but in recent years, robot-assisted minimally invasive esophagectomy (RAMIE) has emerged as a promising alternative. However, robot-assisted surgery can be challenging for novice surgeons, as they often suffer from a loss of spatial orientation. Computer-aided anatomy recognition holds the promise to improve surgical navigation, but research into this area remains limited. In this study, we created a comprehensive dataset for semantic segmentation in RAMIE, encompassing the largest collection of anatomical structures and surgical instruments to date. To determine the most effective method on our new dataset, we benchmarked eight deep learning models using two pretraining datasets. The benchmark includes our RAMIE dataset and the publicly available CholecSeg8k dataset, serving as a general benchmark in surgical segmentation. We found that pretraining on ADE20k is favorable, and attention-based models outperform traditional convolutional neural networks.
Break
Coffee Break 3:00 PM - 3:30 PM
Session 9:
Cancer Interventions
19 February 2025 • 3:30 PM - 5:30 PM PST | Town & Country D
Session Chair:
Junghoon Lee, Johns Hopkins Univ. (United States)
13408-34
19 February 2025 • 3:30 PM - 3:50 PM PST | Town & Country D
Show Abstract +
Frozen section analysis is the standard of care for head and neck squamous cell carcinoma for intraoperative margin assessment. However, accurately relocating positive margins according to the histopathological image back to the patient's resection site is challenging. Significant mucosa shrinkage makes it difficult to establish correspondence between the resected specimen and the resection bed. We propose using nonrigid registration to correct intraoperative deformation by reconstructing a 3D mesh of the resected specimen before shrinkage occurs. A series of cadaver studies demonstrate that deformation correction improves the alignment between the resected specimen and the resection bed. This improved correspondence could help surgeons relocate the positive margins on the patient's resection site.
13408-35
19 February 2025 • 3:50 PM - 4:10 PM PST | Town & Country D
Show Abstract +
This work explores using generative models to enhance artificial intelligence (AI) methods for diagnosing Advanced Gastric Cancer (AGC) lesions through endoscopy. We circumvent the challenge of acquiring large and balanced datasets by using synthetic images to supplement limited and biased training data. For our experiments, we applied this approach to a unique dataset of textural images collected with our Vision-based Tactile Sensor, HySenSe, which captures only partial images of AGC tumors due to its limited sensing area. We trained a conditional latent diffusion model on this dataset to generate high-quality synthetic images and used these to improve our AGC tumor classification system. The objectives were to determine the amount of synthetic data needed and to compare different data addition strategies during model training. Our findings demonstrate that the AI model’s generalizability improves with a mixture of real and synthetic images, although the effectiveness of data mixing methods varies. Overall, the synthetically-enhanced model shows better performance in classifying tumors, even with partial sensor contact images and the mixed morphology of AGC tumors.
13408-37
19 February 2025 • 4:10 PM - 4:30 PM PST | Town & Country D
Show Abstract +
Pancreatic ductal adenocarcinomas are poorly conspicuous lesions that are challenging to accurately assess on imaging throughout treatment course. We establish a biomechanical image registration workflow that leverages Eshelby inclusions to estimate the transformation strain experienced by the tumor during neoadjuvant chemotherapy. This transformation strain is resolved by isolating the change in mass effect concurrently with a biomechanical model that solves for soft tissue deformations that occur between baseline and restaging imaging. In 25 patients with measured ground truth pathologic response to a well-defined neoadjuvant therapy regimen, Eshelby transformation strain was compared to RECIST v1.1 score and change in serum CA19-9 and CEA as non-invasive markers for therapeutic response. We demonstrate that among these clinically available markers, only the novel measure of Eshelby transformation strain was found to be a significant predictor of pathologic outcome.
13408-38
19 February 2025 • 4:30 PM - 4:50 PM PST | Town & Country D
Show Abstract +
This study proposes an uncertainty-aware segmentation method based on evidential deep learning approach which estimates confidence of its inference at the same time of computing HR-CTV segmentation from MRI. We employed a dual convolution-transformer U-Net as a backbone to compute the probability for segmentation for each voxel, and incorporated an uncertainty block to the final output of the network for uncertainty estimation. A total of 250 MRIs obtained from 138 cervical cancer patients, and four sets of manual segmentations were obtained from three radiation oncologists on test dataset to assess the segmentation and uncertainty estimation performance. The proposed model achieved comparable performance to the experts’ manual segmentations. Furthermore, the estimated uncertainty well represented the variability in multi-rater segmentations. These experimental results demonstrate that including uncertainty estimates in HR-CTV segmentation leads to more comprehensive understanding of model prediction, and potentially aids clinicians in making more informed decisions and improving treatment outcome.
13408-39
19 February 2025 • 4:50 PM - 5:10 PM PST | Town & Country D
Show Abstract +
Xerostomia, a common toxicity induced by radiation, severely reduces patients’ quality of life. We propose a deep learning model that predicts the chance of a patient experiencing xerostomia 12 months after radiation therapy. An atlas computed tomography image is created to normalize patient anatomy to help improve model performance. High-resolution class activation maps are generated to better understand the decision of the prediction model. The interpretation of the model’s behavior suggests a correlation between xerostomia and spatial radiation dose in salivary glands.
13408-36
19 February 2025 • 5:10 PM - 5:30 PM PST | Town & Country D
Show Abstract +
A novel method was previously proposed in our earlier work for a matching tool between MRI and spot mammograms. Two registration methods are used for achieving this goal: a biomechanical model based registration between MRI and full X-ray mammograms, followed by an image based registration between full and spot mammograms. In this paper, we focus on evaluating the accumulative registration of the two methods using 51 patients from the Medical University
of Vienna. We also simulate and assess whether the biopsy needle hits the predicted position of the lesion. The
median TRE achieved is 35.6 mm. For 11 or 14 of the 51 datasets an X-ray guided biopsy would be successful
when extracting 12 or 24 specimens with a standard needle biopsy, respectively.
Thursday Morning Keynotes
20 February 2025 • 8:30 AM - 10:00 AM PST | Town & Country B/C
Session Chairs:
Susan M. Astley, The Univ. of Manchester (United Kingdom), Andrzej Krol, SUNY Upstate Medical Univ. (United States)
View Full Details: spie.org/thursday-morning-keynotes
8:30 AM - 8:35 AM:
Welcome and introduction
8:35 AM - 8:40 AM:
Award announcements
- Robert F. Wagner Award finalists for conferences 13407 and 13410
- Computer-Aided Diagnosis Best Paper Award
13407-508
Pioneering vision: The journey of building AI for transformative healthcare
(Keynote Presentation)
20 February 2025 • 8:40 AM - 9:20 AM PST | Town & Country B/C
13410-509
Machine learning in neuroimaging: Understanding heterogeneity of neurologic diseases and building personalized imaging-AI biomarkers
(Keynote Presentation)
20 February 2025 • 9:20 AM - 10:00 AM PST | Town & Country B/C
Show Abstract +
Machine learning has transformed medical imaging in general, and neuroimaging in particular, during the past two decades. We review our work in this field, starting with early contributions on developing personalized predictive markers of brain change in aging and Alzheimer’s Disease, and moving to recent weakly-supervised deep learning methods, aiming to dissect heterogeneity of brain change in neurodegenerative and neuropsychiatric diseases, as well as in brain cancer. We show that disease-related brain changes can follow multiple trajectories and patterns, which have distinct clinical and genetic correlates, thereby suggesting a dimensional approach to capturing brain phenotypes, using machine learning methods.
Break
Coffee Break 10:00 AM - 10:30 AM
Session 10:
Image-Guided Liver Interventions
20 February 2025 • 10:30 AM - 12:10 PM PST | Town & Country D
Session Chairs:
Michael I. Miga, Vanderbilt Univ. (United States), Terry Yoo, The Univ. of Maine (United States)
13408-40
20 February 2025 • 10:30 AM - 10:50 AM PST | Town & Country D
Show Abstract +
This study investigated optimizing antenna insertion pathways for microwave ablation (MWA) to effectively treat liver cancers, targeting a crucial minimal ablative margin of at least 5 mm to prevent tumor recurrence and progression. Utilizing patient-specific 3D models from the Phase II COVER-ALL trials, simulations were conducted to address procedural challenges such as tumor morphology and location. The validated models showed a significant match between predicted and clinical ablation zones. Results from ten patients demonstrated that the optimized antenna insertion pathways can potentially reduce liver tissue damage, ranging from 4.5% to 44% while ensuring effective ablation outcomes. These findings highlight the importance of personalized planning and suggest that tailored MWA antenna insertion based on 3D simulations could standardize and enhance liver cancer treatment outcomes.
13408-41
20 February 2025 • 10:50 AM - 11:10 AM PST | Town & Country D
Show Abstract +
Liver cancer is often treated with surgical resection, but minimally invasive radiofrequency ablation (RFA) is an alternative for unresectable tumors. RFA has a higher recurrence rate due to challenges in monitoring ablation boundaries. This study investigates the use of spectroscopic photoacoustic (sPA) imaging with a customized diffusing fiber to improve visualization of ablation-induced lesions. To address artifacts from sidelobe interference, we implemented the standard Hough Transform (SHT), enhancing necrotic mapping accuracy. Testing on a swine cadaver model showed that sPA imaging effectively identified ablation-induced necrosis, with lesion measurements closely matching gross pathology. These findings suggest that PA-guided ablation could improve precision and outcomes in liver RFA procedures.
13408-42
20 February 2025 • 11:10 AM - 11:30 AM PST | Town & Country D
Show Abstract +
Percutaneous thermal ablation (TA) is a minimally invasive treatment option for select liver tumors. Stereotactic guidance for percutaneous TA depends on reliable respiratory motion control by the anesthesia team to maintain a consistent liver location and shape throughout the procedure for image acquisition, fusion, and ablation probe placements. This is a single-center prospective registry-based study of patients undergoing stereotactic percutaneous TA between 6/2023 and 8/2024. We compare lateral targeting errors between subsequently introduced respiratory motion control techniques and investigated the influence of body mass index (BMI). In addition, we compare overall organ motion and deformation between pre- and post-ablation CT imaging using biomechanical deformable image registration. High frequency jet ventilation (HFJV) resulted in higher targeting accuracy and reduced the influence of BMI compared to breath holds but overall organ deformation was not significantly reduced. HFJV is the preferred option for respiratory motion control during stereotactic percutaneous TA of liver tumors but deformable image registration is still necessary for accurate image fusion.
13408-43
20 February 2025 • 11:30 AM - 11:50 AM PST | Town & Country D
Show Abstract +
Surgical simulations are vital in medical education due to increasing procedure complexity and training costs. This study enhances the realism of laparoscopic cholecystectomy simulations by integrating accurate liver tissue material properties and gravitational effects. Ex-vivo pig liver experiments were conducted to gather data on tissue deformation and force responses to appropriately tune a surgical simulation based on a finite element model with nonlinear material propreties. Using our ex-vivo system experimental force measurements ranged from 1.5 to 2.5 Newtons, which correlated with existing literature. A careful process was created to include gravitational effects into the simulation environment to ensure the simulation accurately reflects force distributions and tissue deformation, thus providing accurate haptic feedback. This improved model offers a more realistic training experience, demonstrating that detailed material characterization and matching appropriate forcing conditions are critical components for effective surgical simulators, ultimately enhancing surgical training and patient outcomes.
13408-44
20 February 2025 • 11:50 AM - 12:10 PM PST | Town & Country D
Show Abstract +
Percutaneous liver tumour ablation is the preferred treatment for patients ineligible for surgery. In this procedure, an accurate needle placement is crucial for success. While Ultrasound(US)-guided manual insertion can achieve tumour access, it often requires multiple needle repositioning, increasing the risk of patient harm. Surgical Navigation Systems (SNS) are employed to provide additional information and mechanical support, to the surgeon. Our team has developed a Mini-SNS that combines a mini stereotactic, patient-attached, needle guider with virtual reality (VR) to facilitate needle positioning. In this work, we present a preliminary user trial comparing conventional US needle guidance puncture with our Mini-SNS. Preparation time, insertion time, total procedural time, needle repositions, and final needle (real and virtual) tip positions were recorded and analyzed. These initial findings suggest that the Mini-SNS enhances efficiency in surgical navigation, with reduced insertion time and repositioning, supporting its potential to minimize patient harm.
Program Committee
The Univ. of Texas at Dallas (United States), The Univ. of Texas Southwestern Medical Ctr. (United States)
Program Committee
National Institute of Allergy and Infectious Diseases (United States)