Rapid advances in imaging, sensing, and their underlying mathematics are impacting numerous areas of medicine, with promising progress in heart surgery, cancer detection and treatment, tablet-controlled microsurgery, cochlear implants, computer-aided diagnosis. Talks reported during SPIE Medical Imaging 2015, 21-26 February, in Orlando, Florida, described some of the most promising research.
Real-time imaging and image fusion help heart procedures
Treating cardiac arrhythmias through ablation, or scarring of heart tissue, presents clinicians with many challenges. Ablation can be a lengthy surgery -- as long as 12 hours -- and visualizing defective areas of a beating heart in motion can be difficult. Accurately visualizing the area in real time means a more effective surgery and is necessary for guiding catheters used in intracardiac ablation from their insertion in the groin to the heart.
During the plenary session, the Mayo Clinic's Douglas Packer presented his work (video), in which he uses multiple imaging modalities for guiding intracardiac and noninvasive extracorporeal ablation, finding that combining real-time imaging and fusing modalities provides the best results.
Pre-procedure CT/MR imaging, for example, provides global views of the anatomy while ultrasound offers a localized view of portions of the region. Accurate registration during overlay of these images is critical to effective diagnosis and treatment, says Packer. This is especially true for navigating catheter paths through the heart, and in using strain and voltage measurements to locate irregularities.
For extracorporeal ablation, Packer's investigations using x-ray radiation and particle beams have recently focused on the use of carbon ions to deliver energy to the ablation target.
Automated tools for cancer diagnosis
The Digital Pathology conference featured a session on gastro-intestinal and genito-urinary cancers, which revealed steady progress toward automated tools for more effective diagnosis and clinical decisions.
Geert Litjens of Radboud University, Nijmegen Medical Center, demonstrated an automated system for grading and analyzing digitized whole slide images for prostate cancer diagnosis, while Case Western researcher Asha Singanamalli showed that quantitative features of gland morphology, architecture, and orientation can be computed from 7 Tesla ex vivo MRI. Singanamalli's technique is also able to distinguish between cancerous and benign tissue in the prostate by detecting differences in co-occurring gland tensor features.
Marios Gavrielides of the U.S. FDA presented work on an ovarian cancer observer study concluding that observer performance is subtype specific; and Faisal Kahn, Icahn School of Medicine at Mount Sinai, showed that protein biomarkers and immunofluorescent images combined with glandular morphometric features performed better when used together than alone as predictors in prostate cancer.
Oscar Geessink of the University of Twente concluded the session by presenting an automated tool for objective quantification of stroma and tumor proportions in studying colorectal cancer, which demonstrated less variability than human observers.
Compressed sensing accelerates medical imaging advances
Medical imaging has advanced rapidly in the past 10 years, in particular in the area of compressed sensing. Daniel Sodickson, New York University School of Medicine, presented a fascinating keynote presentation on the role of compressed sensing (see video), the advances it has enabled, and the promise it holds for imaging in the future, as part of the Biomedical Applications in Molecular, Structural, and Functional Imaging conference.
While fundamental concepts of tomographic imaging have been constant since CT and MRI came on the scene in the 1970s, the applied mathematics used to collect and reconstruct the data have advanced significantly. Compressed sensing allows the reconstruction of images with less data than other techniques require, and while there are limits to the approach, the results are powerful.
Sodickson showed a 20-fold acceleration in an abdominal scan providing full liver coverage using a continuous imaging compressed sampling method known as GRASP (golden angle radial sparse parallel MRI). Developed by Sodickson and collaborators, GRASP has been used on more than 3,000 patients and has been developed further into XD-GRASP to decrease motion-related artifacts.
He then combined these imaging techniques with rapid compressive imaging techniques that use long pulse sequences, resulting in 3D radial scanning that enables free-running sequences with multiple views for imaging applications such as cardiac MRIs. These concepts hold great promise for further advances in medical imaging.
Improving patient experience and robotic systems
Surgical strategies and improving patient experience were the focus in the Treatment Planning and Robotic Systems session during the Image-Guided Procedures, Robotic Interventions, and Modeling conference.
Andreas Schoob of Leibniz University Hannover presented work on a tablet-controlled prototype laser surgery system. His investigation included methods for incision path definition, including continuous and point-based schemes. Schoob's team found that pen-based and pen display routines provide the best results, and they plan to extend the study to three-dimensional paths.
Yiyuan Zhao, Vanderbilt University, presented work on improving cochlear implants for the hearing impaired. The Vanderbilt group has developed an image-guided, automated way to determine which electrodes to activate once the device has been implanted, which is key to optimizing the implant's ability to provide sound signals to the brain.
Work on micro-stereotactic frames includes the first steps toward a customized and disposable assembly for minimally invasive ear surgeries, particularly those that require deep brain stimulation. Lüder Kars of Leibniz University Hannover described a new method for such frames using 3D-printed plates and the use of a bone cement to set the fixture position while the structure is in the desired position on the patient.
Additional presentations in the session included a nonholonomic catheter path reconstruction for treating prostate cancer from researchers at Queen's University in Canada, navigation tools for brachytherapy from the Canadian Surgical Technologies and Advanced Robotics (CSTAR) center, and methods to non-invasively generate patient-specific pulmonary vascular models to improve surgery and decrease the need for exploratory surgeries from a team at the University of Florida.
Next generation of clinical decisions will benefit from computer-aided diagnosis
Tanveer Syeda-Mahmood of IBM Almaden Research Center delivered a keynote presentation (see video) on advances in machine learning for clinical decision support (CDS) at the Computer-Aided Diagnosis conference. CDS has benefited recently from the combination of healthcare data, interference algorithms from artificial intelligence, machine-learning techniques, and patient similarity. Syeda-Mahmood says this multimodal approach provides exciting opportunities and unique challenges for the computer-aided diagnosis community.
She reviewed the progress of these different aspects of CDS and says the next generation of CDS will involve cognitive assistants using multimodal reasoning and machine learning in all stages of disease detection, which will greatly benefit healthcare in the future.
Challenges include combining electronic health record data with clinical knowledge, generating and labeling large-scale datasets, and adding pathologic and genomic data for analysis. Benchmarking and comparison of the performances of various algorithms will be needed to improve the process, she said, in preparation for FDA approvals and clinician acceptance and adoption.
Conference proceedings are being published online in the SPIE Digital Library as approved, with CD and print publication to follow.