• About the Society
  • SPIE Altruism
  • Advocacy
  • International Day of Light
  • Awards Programs
  • Press Room
  • Press Releases
    SPIE Member News
    Event News
    Media Contacts
    Press Registration
    SPIE Logos and Name
  • Jobs at SPIE
Print PageEmail Page

SPIE Medical Imaging 2015 news

 

SPIE Medical Imaging, Orlando

Plenary talk: Multiple imaging modalities in heart procedures

LUNGx: the CAD Grand Challenge

Keynote talks

Harnessing big data to improve healthcare

Merging modalities to improve imaging schemes

Machine learning for clinical decision support

Target registration errors

Biomarkers in cancer treatment

Open-source platforms for managing data

Connecting image perception with radiology for more effective breast screening

Glimmerings of a unified tomography

Pathology tools for image-based search

Technical sessions

Treatment planning and robotic systems

Digital pathology: gastro- cancers

Visual search in image perception

Imaging the lung

Registration systems and tools

Evening workshops and panel discussions

Student lunch with the Experts

Awards

Poster receptions

Medical imaging and the International Year of Light

Time to network!

 


Plenary talk: Multiple imaging modalities in heart procedures

Douglas PackerMonday's plenary session at SPIE Medical Imaging 2015 featured an inspiring talk by Douglas Packer (at right) of the Mayo Clinic. Packer presented his translational work on the use of multiple imaging modalities for guiding intracardiac and extracorporeal ablation for the treatment of cardiac arrhythmias.

The presentation gave conference attendees a glimpse into the applications of imaging modalities featured at this week's forums.

Packer noted issues with current procedures ranging from extended operating time (as long as 12 hours in some cases) to the tendency to expand the ablation region when the defective area is difficult to visualize. Medical procedures are further complicated by the fact that the heart is beating and therefore in motion during the procedure. Clinicians have a need to accurately visualize the area in real-time to most effectively ablate the targeted region, he noted.

Pre-procedure CT/MR imaging provides global views of the anatomy under study while ultrasound offers a localized view of portions of this region, implying that accurate registration during overlay of these images is critical to effective diagnosis and treatment, he said. This is particularly true for intracardiac procedures where the images are used to navigate catheter paths through the heart.

Packer noted that imaging extends beyond three dimensions in a sense, given both the time component and the use of strain and voltage measurements to locate irregularities.

In terms of extracorporeal ablation, investigations using x-ray radiation and particle beams have begun and have focused lately on the use of carbon ions to deliver energy to the ablation target, Packer said.

Packer's presentation featured a number of examples of image fusion in the cases of human and animal models. The complexities of real-time accurate imaging and fusion of modalities provided the audience with deeper insight into the use of the technology they are developing and should serve to inspire the medical-imaging community with new challenges and opportunities.

 


LUNGx: the CAD Grand Challenge

A Grand Challenge on developing quantitative image analysis methods for the diagnostic classification of malignant and benign lung nodules culminated in a panel discussion and awards Tuesday.

Conducted by SPIE, AAPM, and the U.S National Cancer Institute (NCI), the challenge provided an opportunity for participants to compare their algorithms to those of others from academia, industry, and government in a structured, direct way using the same data sets -- and in the process, play a vital role in the selection of promising classes of algorithms and systems for further clinical translational efforts. The goal is to prompt advances in computer-aided diagnosis and ultimately precision medicine.

Lyndsey Pickup, Mirada Medical UK, was selected as the winner, and Yoganand Balagurunathan, Moffitt Cancer Center, University of Arizona was named runner-up.

Discussion covered current and future opportunities for CAD Grand Challenges as a testbed and in enabling cross-platform decision support system evaluation.

LUNGx CAD Grand Challenge at SPIE Medical Imaging

From left, panelists and winners included (front row) Stephen Aylward of Kitware, Inc., George Redmond of the NCI, Laurence Clarke of the NCI, winner Lyndsey Pickup of Mirada Medical, Karen Drukker of the University of Chicago Medical Center, and Samuel Armato, The Univ. of Chicago; and (back row) runner-up from the University of Arizona, Nicholas Petrick of the U.S. Food and Drug Administration, and runner-up Yoganand Balagurunathan of the University of Arizona.

 


Keynote talks

 

Harnessing big data to improve healthcare

photo of Dr. SiegelHarnessing big data to potentially and significantly alter and improve the field of medical imaging and its role in healthcare was the topic of a keynote presentation by Eliot Siegel, professor and vice chair at the University of Maryland School of Medicine and Computer Science.

In the "PACS and Imaging Informatics: Next Generation and Innovations" conference, Siegel discussed the need to improve image data collection and sharing as well as the need for new statistical methods to analyze the breadth of data being collected.

The use of all this data for computer-aided diagnosis (CAD) has plateaued, he said. Even in mammography, where it is most used, Siegel cited survey findings that only 2% of clinicians stated that they always rely upon CAD results and 62% reported that they rarely change a diagnosis based on CAD results.

The offset between use and reliance in mammography leads one to ask if computer assistance is truly valuable. Siegel's examples showed that image perception and observer performance can bias a diagnosis. He discussed the complementary nature and, therefore, benefit of computers and human observers in making diagnoses.

Personalized and precision medicine provide opportunity to collect and share data for the patient's good. Researchers in large studies can collect baseline statistics for various demographic groups for use in making diagnoses. Siegel pointed out, however, that only 2-3% of cancer patients are in clinical studies, which means that the data for 97% of patients is not recorded for larger consumption and use. Securely consolidating this type of data is one opportunity for the community.

Noting that images provide a large dimension of data, Siegel observed that many images contain more data than that for which the image was originally recorded. Being able to store and access those images for future study would prove beneficial.

Siegel cited a number of challenges in effectively and efficiently sharing images including acquisition standards, annotation standards, the use of a common lexicon, variations in software platforms in image interpretation and the lack of reference imaging sets.

The amount of data collected is another challenge, namely, the need for more sophisticated statistical analysis tools. Methods such as latent variable analysis and principal component analysis, familiar to those in medical imaging community, now need to be mapped to analytical schemes.

 

Merging modalities to improve imaging schemes

photo of Dr. AnastasioMerging modalities to produce improved imaging schemes benefiting from the strengths of the individual methods is an evolving theme in medical imaging. SPIE Member Mark Anastasio from Washington University in St. Louis (USA) discussed the combination of photoacoustic computed tomography (PACT) and ultrasound computed tomography (USCT in his keynote presentation for the "Ultrasonic Imaging and Tomography" conference.

PACT is a relatively new modality comprised of a hybrid scheme in which light is introduced and absorbed by a sample, resulting in the excitation of acoustic waves subsequently detected by transducers. The method provides both strong image contrast and a spatial resolution similar to what is achieved in ultrasound. The intrinsic heterogeneity of most systems results in a variation in the speed of sound (SOS) which can, in turn, result in imaging artifacts. Including USCT methods allows for estimates of the SOS variations that can then be added to the PACT data to improve image quality. In addition, the UCST data can be overlaid with the PACT data to further enhance image reconstruction. Utilizing this additive scheme, Anastasio demonstrated results with a living mouse model.

Noting that PACT data also contains SOS information, Anastasio then described a new approach exploiting all PACT data and using it with USCT data in an integrated reconstruction scheme. This method has the advantage of requiring less UCST data, and, hence, scans, to provide accurate, high-contrast images.

 

Machine learning for clinical decision support

Tanveer Syeda-MahmoodTanveer Syeda-Mahmood of IBM Almaden Research Center delivered a keynote presentation on advances in machine learning for clinical decision support (CDS) in the Computer-Aided Diagnosis conference. The confluence of healthcare data with inference algorithms from artificial intelligence, machining-learning techniques, and patient similarity is driving a multimodal approach for CDS, she said.

The age of big data coupled with the expansion of modalities provides exciting opportunities and unique challenges for the computer-aided diagnosis community, she said. Embracing the opportunities and overcoming the challenges promises to take the field of CDS to a new level of increased value in the healthcare field.

Syeda-Mahmood noted that inference-based CDS has evolved through semantic networks and knowledge graphs to decision trees, Bayesian networks, and neural networks. Machine learning-based CDS has utilized clustering techniques, classification methods, and support vector machines, ensemble learning, and the so-called "random forest" methods.

In machine learning today, deep learning techniques utilizing neural networks are enjoying a revitalization due to the onset of fast GPU's capable of efficiently exploring parallel paths. Patient similarity, brought to the forefront over the past several years by such efforts as the IBM AALIM (Advanced Analytics for Information Management) system, is benefiting from methods for automatic distance metric learning and becoming multimodal in scope.

The next generation for CDS, she said, will involve cognitive assistants using multimodal reasoning and machine learning in all stages of disease detection. Diagnostic interpretation of modalities will be an area of interest, Syeda-Mahmood said.

Challenges include combining electronic health record (EHR) 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.

 

Target registration error

Michael Fitzpatrick, SPIE Medical ImagingTarget registration error, referred to as TRE, is a term and quantity familiar to those in the image registration and surgical guidance communities. But that wasn't always the case, according to Michael Fitzpatrick of Vanderbilt University.

The story of the evolution of TRE and its associated quantities, fiducial registration error (FRE), and fiducial localization error (FLE), was the subject of Fitzpatrick's entertaining and sometimes humorous keynote in the "Image-Guided Procedures, Robotic Interventions and Modeling" conference.

Fitzpatrick detailed the history of the work that led to the equation for TRE, starting with initial studies in 1986 by Dr. David Roberts at Dartmouth and followed closely by the work of his own group and collaborators at Vanderbilt.

Efforts began when Fitzpatrick was approached by George Allen, head of Neurological Surgery at Vanderbilt, about replacing the bulky head-mounted frames patients wore in brain surgeries with a series of markers. These markers would require a registration scheme for mapping pre-surgery CT data.

Intertining the stories of the development of the surgical system with the research into TRE illustrated the pitfalls and progress in both industrial and academic settings. The historical review also demonstrated that setbacks and poor reviews by editors and funding agencies can ultimately be overcome through perseverance, dedication and a bit of luck.

For example, Fitzpatrick noted that an early sponsor of the work, J&J, killed the project just hours prior to shipment of the first system in December of 1995, thereby removing themselves from a multibillion dollar a year industry. It was that same year that Fitzpatrick drew inspiration from quantum mechanics and perturbation theory to drive the mathematical framework that produced the equation for TRE first presented at an SPIE conference in 1998.

 

Biomarkers in cancer treatment

photo of Dr. MankoffThe use of biomarkers in cancer treatment employs tracers to detect the presence or absence of tumors. Thinking of a different role for imaging as a means to guide cancer treatment selection and match the therapy to the tumor biology was the subject of Tuesday's keynote presentation in the ‘Physics of Medical Imaging' conference delivered by David Mankoff of the University of Pennsylvania.

The emerging paradigm for imaging is to use biomarkers to measure the factors impacting response to enable effective treatment, he said. Choosing the right patient, choosing the right drug, getting the right result, and predicting outcome are the four key elements to this new approach, said Mankoff.

In terms of choosing the right patient, it is important to determine if the therapeutic target is present, he said. An example targeting breast cancer therapy demonstrated that FES uptake could be used to predict tumor response to hormonal therapy. Choosing the right drug means insuring that the drug reaches its target. He illustrated this with the results of a study utilizing cyclosporine to inhibit P-gp. P-gp is a protein located in cell membranes, which limits drug transport to the brain. Suppressing it with cyclosporine enabled C-verapamil uptake in the target region, Mankoff explained.

Getting the right results means understanding if there is a pharmacodynamic response with imaging techniques designed to study tumor cell proliferation. Mankoff discussed a case study using thymidine, a DNA building block in this context.

Finally, Mankoff illustrated how to predict outcome and understand the correlation of biomarker response to patient survival rates with a study showing FDG PET imaging results collected early in lymphoma treatment, providing indication of patient responsiveness and longevity. Mankoff said that scanner cross-calibration, the development and supply of novel PET probes, and frameworks for trials to validate results are needed to advance the field of molecular imaging.

 

Open-source platforms for managing data

photo of Jason SwedlowIn his keynote, Jason Swedlow of the University of Dundee demonstrated the power of open source platforms for sharing, analyzing, and managing large quantities of data from many medical-imaging modalities.

The collection and use of images is ubiquitous across the life sciences and biomedical research communities today. The images are acquired using many modalities such as electron microscopy, MRI, and fluorescent imaging. Swedlow told attendees at the "Image Processing" conference about the importance of treating each image as a resource containing quantitative data and developing a mechanism to allow the sharing of this information.

One measure of success for such an open environment is the ability of images from a new modality to access and use analytical tools already in place. Towards that end, the field needs an effective file translator and data-management tool, Swedlow said.

The Open Microscopy Environment, founded by Swedlow in 2000, is a consortium of researchers in Europe and the U.S. building and supporting such tools. Bio-Formats is a Java-based library capable of reading multiple file formats, and OMERO is client-server software enabling image management, visualization, and analysis. Bio-Formats is able to read over 100 proprietary image formats converting them to the OME-TIFF standard while OMERO supports multiple programming languages and platforms. Projects such as FLIMfit, a software tool for the analysis of fluorescence lifetime imaging data, utilize OMERO.

The consortium hopes to address more data types and enable 3D images in particular. They are also interested in finding ways to handle the expanding amount of metadata in images and enhance image publishing.

 

Connecting image perception with radiology for more effective breast screening

Diane Georgian-Smith at SPIE Medical ImagingBridging the gap between communities to develop better solutions was the theme of the keynote talk for the Image Perception, Observer Performance, and Technology Assessment conference delivered by Diane Georgian-Smith of Harvard Medical School and Brigham and Women's Hospital.

In particular, bridging the gap between the image perception and radiology communities to enable more effective breast screening was the focus of presentation.

Georgian-Smith began by sharing a statement in a recent doctoral thesis she reviewed where the candidate noted that over the past 30 years the miss rate on lung lesions has not changed. Clearly this represents an opportunity to improve and working together perhaps these two communities can develop better screening tools.

Towards that end, the speaker sought to educate the audience on what radiologists look for when viewing images and how they process what they see in these images.

Radiologists look for three things: mass, calcification and architectural distortion. Their analysis and ability to process images is hampered by the fact that they work with only five shades of gray-related to air, fat, soft tissue, calcium, and metal.

Understanding mass entails looking first at the shape and then the margins to determine if the feature is benign or malignant. Calcification involves looking at morphology and distribution. Architectural distortion looks for straight lines, a certain sign something has changed in the body, and radiated features.

In the case of lesion detection, mass brightness is often less important than are the mass features. Methods such as tomosynthesis improve detection and help in the interpretation of architectural distortions but even this method has limits.

There is also the question of being able to effectively teach the principles of reading mammograms and detecting malignancies.

A three-phase study with radiologists seeking to improve detection and diagnosis related to architectural distortions demonstrated increased specificity suggesting that certain aspects can be effectively taught and learned.

In terms of mapping diagnosis to computers, the results of a second reader study where the second reader was either a CAD reader or a human reader showed clear differences in the recall rate between the two review methods as well as differences in detection and false readings indicating a challenge, but, perhaps, more importantly, an opportunity for the two communities to work together to both close this gap and improve the accuracy of a diagnosis.

Working with one another, the communities can bring their complementary skills together to make a better radiologist.

 

Glimmerings of a unified tomography

Daniel Sodickson at SPIE Medical ImagingThe fundamental concepts of tomographic imaging have remained relatively constant since the onset of CT and MRI in the 1970s. However, rapid advances in applied mathematics in the form of compressed sensing techniques over the past decade have provided new ways for the imaging community to collect and reconstruct data.

The role of compressed sensing, the advances it has enabled, and the promise it holds for imaging in the future was the subject of the keynote presentation for the Biomedical Applications in Molecular, Structural, and Functional Imaging conference provided by Daniel Sodickson of the New York University School of Medicine.

The past decade has been a time of accelerated learning and capability in the field of medical imaging with advances in dynamic imaging enabled by the principles of compressive sensing. The concept depends upon the sparsity of data coupled with incoherent sampling to reconstruct images using less data than what the Nyquist theorem would require.

There are limits to the approach. For example the maximum acceleration depends upon the underlying degree of sparsity. Nonetheless, examples showing 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) demonstrated the power of the technique.

GRASP, developed by the author and his collaborators, has been used on over 3,000 patients, but the method does show residual motion-related artifacts. This led the team to improve the technique by examining motion sorting schemes as opposed to motion correction methods. XD-GRASP (where 'XD' means 'extra dimension') enables multiple images through motion using the same data but sorting the data differently.

In addition to, and complementing, the principles of continuous imaging exploited by techniques like GRASP and XD-GRASP are rapid compressive imaging techniques utilizing long pulse sequences. Combining all these concepts into a method named 5D-XD-GRASP using 3D radial scanning enables free-running sequences with multiple views as demonstrated in an example involving a cardiac MRI. Multimodal imaging is also possible with these techniques and promises to provide more information for the clinicians on which to base diagnoses.

This exciting keynote presentation highlighted the rapid rate of advance in medical imaging enabled by new mathematics which holds the promise of providing further improvements and greater acceleration in the field. As the author suggested, we may be entering a new world of medical imaging.

 

Pathology tools for image-based search

Ulysses Balis at SPIE Medical ImagingEnabling pathologists with the tools of the digital age was the message in the Digital Pathology conference keynote talk delivered by Ulysses Balis of the University of Michigan, who stated early in the presentation that the field is something of an art in need of decision support.

The presentation discussed imaging in pathology, compared pathology to the more digital-technology-advanced field of radiology, discussed some of the challenges unique to pathology, described some of the opportunities for the imaging community in engaging with pathologists and touched on some of the tools emerging today.

Image-based search, also known as content-based image retrieval (CBIR), is the process of searching image databases with exemplar images. Effective deployment of CBIR will be one important mechanism to the more effective utilization of imagery data. Linking those search results to metadata associated with the image will make for a powerful tool for pathologists.

For pathology, whole slide imaging began approximately 25 years ago with researchers cobbling together lab systems. Today, commercially available systems complete with features such as autofocus are available but typically used only in selective cases and not the full work flow.

Taking this to the next level requires better standards and better tools for the management and exchange of large images and image libraries. Having this and being able to search a repository for matches and extract metadata detailing information such as prior diagnoses and biological potential of malignancy would give pathologists a powerful decision support tool.

Radiology embraced digital content in the 1980s and by the mid-2000s was essentially all digital. Pathology has a few isolated labs on the verge of being all digital and hence trails its counterpart by a decade or more in this respect. Whereas radiology data is largely digital, has algorithms and phantoms enabling standardization, and has physical samples and instrumentation that is tightly controlled, pathology data is largely analog, has little standardization, and has samples and dyes which are sensitive to the environment. Pathology also works across a range of length scales not typically seen in other disciplines.

The age of big data presents a number of challenges and opportunities for managing images and databases. Blending orthogonal data tied to an image and being able to efficiently search and extract it will be crucial to establishing a successful system.

One can adopt either an information-theory or domain-specific approach to addressing this challenge. Again, standards, and ground truths will be required. Despite these challenges, some organizations are on the vanguard of engaging in this opportunity. The presentation include a demonstration of an image analysis routine followed by a database search which linked results to metadata.

The presentation illustrated some of the challenges in moving a traditionally qualitatively centered field into areas of quantitative analysis and the digital era. These are exciting challenges for the imaging community to embrace particularly given the importance of successful field deployment of the type of capability Balis described.

 


Technical sessions

 

Treatment planning and robotic systems

Treatment Planning and Robotic Systems was the topic of a Sunday afternoon session in the conference on Image-Guided Procedures, Robotic Interventions, and Modeling.

Examining tablet-based strategies for laser microsurgery was the focus of a presentation by Andreas Schoob of Leibniz University Hannover. Investigating laser surgery approaches for vocal fold tumors, the authors constructed a tablet-controlled prototype laser surgery system to investigate various methods for incision path definition.

Methods studied included various continuous and point-based schemes and revealed that pen-based and pen display routines provided the best results in terms of root-squared mean errors and maximum displacement errors for a series of test patterns.  Future work will include extending the study to three-dimensional paths.

Optimizing the effectiveness of cochlear implants was the topic of the presentation by Yiyuan Zhao of Vanderbilt. Electrodes on the implant are used to stimulate various regions of modiolus. After implant, audiologists assign stimulation settings to the various electrodes to optimize hearing but it is possible to have multiple electrodes stimulating the same regions resulting in a decrease in overall effectiveness of the device.

The authors have developed an image-guided approach coupled with an automated method to determine which electrodes to activate. Using implants from two different manufacturers, the authors reported results that were as good as or judged to be superior to those achieved by trained audiologists in at least eighty percent of the cases tested.  The work provides a pathway to improving the quality of life for the hearing-impaired.

Microstereotactic frames are under study for minimally invasive temporal bone surgeries such as those requiring deep brain stimulation. Lüder Kars of Leibniz University Hannover described a new method for such frames utilizing 3D-printed plates and the use of a bone cement to set the fixture position while the structure is in the desired orientation on the patient.

Initial tests indicate the desired accuracy and stability should be achievable with this platform. This work provides the initial steps towards a customized and disposable assembly to optimize these types of surgical procedures.

Additional presentations in the session included a nonholonomic catheter path reconstruction for use in prostate cancer studies 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 surgical procedures and decrease the need for exploratory surgeries from a team at the University of Florida.

Overall, the session highlighted the value of well-constructed surgical strategies and planning to improve patient experience.

 

Digital pathology: gastro- cancers

Thursday morning's Digital Pathology conference featured a session on gastro-intestinal and genito-urinary cancers. Geert Litjens of Radboud University, Nijmegen Medical Center, demonstrated an automated system for grading and analyzing digitized whole slide images for prostate cancer diagnosis.

Prostate cancer was the subject of the following talk as well, as Asha Singanamalli of Case Western discussed work showing that quantitative features of gland morphology, architecture and orientation can be computed from 7 Tesla ex vivo MRI and that the technique is able to detect differences in co-occurring gland tensor features between cancerous and benign tissue.

Marios Gavrielides of the U.S. FDA presented work on an ovarian cancer observer study concluding that observer performance is subtype specific.

The use of protein biomarkers and immunofluorescent images used in conjunction with glandular morphometric features was shown to outperform the use of these techniques as individual predictors in prostate cancer analysis in a presentation by Faisal Kahn of the Icahn School of Medicine at Mount Sinai.

Oscar Geessink of the University of Twente concluded the session with his presentation of an automated tool for objective quantification of stroma and tumor proportions in the study of colorectal cancer. The algorithm developed and tested demonstrated less variability with respect to ground truth images than did the results of human observers.

This session featured excellent presentations and question/answer sessions with the audience showing that the digital pathology community is making steady progress towards the development of automated tools for more effective diagnosis and decision support.

 

Visual search in image perception

A morning session of the Image Perception, Observer Performance, and Technology Assessment conference dealt with the topic of visual search.

Noting that prevalence has a strong effect on how people make decisions, Frank Samuelson of the U.S. FDA described work building a model of observer behavior based on linear decision theory and Bayesian learning principles that fit data related to frequencies of false negatives and false positives (FNF and FPF) well in the test cases studied.

Anando Sen of the University of Houston studied how model observers respond to sources of variation in comparison to human observers using channelized Hotelling observer, channelized non-prewhitening observer, and visual observer models, and concluded that the visual observer model was the closest to humans in terms of localization and offers a computationally efficient method to account for anatomical noise.

In a study with radiologists reading a set of forty mammograms, Sarah Lewis of the University of Sydney reported that priming the set with obvious cancers may disturb the visual perception and search pattern of the observers.

Folami Alamudun of Texas A&M University utilized fractal analysis to study scanning patterns of radiologists and reported that case pathology, image density and observer experience are independent predictors of gaze complexity. Gaze behavior is drawing attention in security communities for use as a biometric.

At last year's meeting, Georgia Tourassi of Oak Ridge National Laboratory and coworkers reported on results indicating that gaze velocity depends upon the nature of the organizational task and, in an extension of this study, noted this year that the temporal stability of this parameter is of concern and may compromise its value for biometrics.

The session concluded with Howard Gifford of the University of Houston sharing early results by his team on the potential of the visual observer model as a human surrogate.

Taken together the talks in the session provided new insights into image perception and observer performance.

 

Imaging the lung

Lung imaging was the subject of the first Thursday afternoon session for the Biomedical Applications in Molecular, Structural, and Functional Imaging conference.

Nanxi Zha of the University of Western Ontario discussed the use of principal component analysis on CT density histograms in the study of COPD.

Fumin Guo of the Robarts Research Institute and coworkers reported on their efforts to develop an automated CT and noble gas MRI registration and segmentation process to provide lobar structure-function measurements to guide pulmonary therapy.

A novel 3D lung cine MRI method using a sweep imaging Fourier transform approach was described by Naoharu Kobayashi of the University of Minnesota.

Computing textural features with a ground glass opacity model for studying radiation-induced lung injury following stereotactic radiotherapy for treating lung cancer patients was the topic of the presentation contributed by Sarah Mattonen of the University of Western Ontario.

Dante Capaldi of the Robarts Institute closed the session reporting on Fourier decomposition pulmonary MRI, a non-invasive MRI method not utilizing contrasting agents, for use in ventilation imaging. The session highlighted the wide range of work being done to progress the field of lung imaging.

 

Registration systems and tools

The final Thursday session of the Image Processing conference addressed the topic of registration.

Marc Modat of University College London studied metal-on-metal hip arthoplasties and resulting hip abductor muscle atrophy using a multi-axis segmentation scheme and demonstrated the method was suitable for use with disease status specific databases.

Reporting on a study of five commercially available registration tools for the difficult problem of abdominal cavity CT registration, Christopher Lee of Vanderbilt University concluded that care must be taken given the complexity of the abdomen when using these tools particularly when it comes to addressing data outliers.

Mazen Alhrishy of King's College London described the use of a novel 2D/3D image registration technique designed to reduce the need for repeated digital subtraction angiography imaging.

Rui Hua of the University of Sheffield then discussed work to extend a free form deformation framework in a computationally efficient fashion to handle motion in non-rigid registration cases.

A method for addressing multi-modal deformable image registration using a higher dimensional registration scheme and estimating missing data was the subject of the talk given by Min Chen of Johns Hopkins University.

The final talk of the session was provided by Tanya Alderliesten of Academisch Medisch Centrum who described a multi-objective optimization scheme that assisted with guidance information in the case of deformable object registration.

The results illustrated the value of the approach for large deformations and disappearing structures. Overall, the talks in this session highlighted the novel image processing approaches being taken to complex registration problems whose solutions are important to the medical imaging community.

 


Evening workshops and panel discussions

SPIE Medical Imaging audience

Sunday evening featured a series of conference workshops and panel discussions including one chaired by Robert Webster of Vanderbilt as part of the conference on Image-Guided Procedures, Robotic Interventions, and Modeling. The workshop, entitled Novel Robots for Less Invasive Surgery, featured presentations on emerging robotic surgery methods by four researchers followed by a brief panel discussion.

Shing Shin Cheng of the University of Maryland described the use of shape memory alloys (SMA) in the design and build of a neurosurgical intracranial robot.  The use of SMA materials provides the necessary dexterity for maneuvering while maintaining strength.  To control thermal effects, a water cooling system was integrated into the system as well.

The design and use of continuous robots comprised of concentric NiTi tubes was discussed by Jessica Burgner-Kahrs from Leibniz University Hannover.  Robots have been designed, built and tested for at least three different applications: transnasal skull base surgery, intracerebral hemorrhage evacuation, and atraumatic cochlear implant insertion. The work demonstrated impressive initial results and points to the promise of the technique.

The theme of concentric tube robots was further explored by Sung-Chul Kang of the Korea Institute of Science and Technology who discussed the use of three nitinol tubes in a robot equipped with forceps for endonasal skull base surgery. This system is in preclinical, cadaver-based tests. Kang also described the Dr. Hujoon Project. Named after the first Korean physician, from the 16th century, the project is developing a microsurgical robot based on an open platform.

The final speaker of the evening, Gregory Fischer of Worcester Polytechnic Institute, discussed intraoperative MRI and enabling closed loop interventions. MRI guided stereotactic neurosurgery and MRI-guided prostate cancer diagnostics and therapy were two of the applications discussed, and the presentation highlighted a number of innovations for imaging, registration, and tracking  to optimize MRI-based processes.

The panel discussion featured several questions from the audience including one on the challenges bringing these novel technologies to markets. Several of the panelists agreed that feedback, control and integration of imaging and sensing capabilities was crucial to advancing the field. Pulling together this community with members of the sensing and biophotonics communities would be a good next step.

 


Student Lunch with the Experts

MI15 Student Lunch with the Experts

Symposium Chair David Manning welcomed students from around the world to the SPIE Student Lunch with the Experts. Future leaders in the optics and photonics community dined with experts in the field who were willing to share their experience and wisdom on career paths.

SPIE Medical Imaging Student Lunch with the Experts

SPIE Medical Imaging Student Lunch with the Experts

 


Awards 

MI15 Scholarship recipients

SPIE Scholarship recipients (left to right) Ning Yu (University of Pennsylvania Health Systems), Parmida Beigi (University of British Columbia), and Alex Dawson-Elli (Rochester Institute of Technology) received their certificates at the Student Lunch with the Experts.

Ibrahim Sadek, Steven Horii at SPIE Medical Imaging

Ibrahim Sadek (Université de Bourgogne) was named the Robert F. Wagner Best Student Paper First-Place Winner, for "Automatic discrimination of color retinal images using the bag of words approach" (9414-54). Above, Sadek's co-author Fabrice Meriandeau accepts the award from symposium chair Steven Horii (University of Pennsylvania Health System).

 Fumin Guo, Robarts Research Institute with Steven Horii at SPIE Medical Imaging

Fumin Guo (Robarts Research Institute), at left, was named the Robert F. Wagner Best Student Paper Award Runner Up for "Automated pulmonary lobar ventilation measurements using volume-matched thoracic CT and MRI" (9414-42).

Robert F. Wagner Best Paper finalists

The Robert F. Wagner Best Student Paper Award finalists.

Names of more winning oral and poster authors will be posted soon.

 


Poster receptions

In addition to oral presentations, plenary talks, and technical workshops, SPIE Medical Imaging highlights included poster receptions on Monday and Wednesday evenings.

photo from poster session, SPIE Medical Imaging 2015

Robert Nishikawa and others at poster session, SPIE Medical Imaging 2015

Thomas Deserno and others at poster session, SPIE Medical Imaging 2015

photo from poster session, SPIE Medical Imaging 2015

photo from poster session, SPIE Medical Imaging 2015

 


Medical imaging and the International Year of Light

International Year of Light at SPIE Medical Imaging

Marie Curie, Chandrasekhara Raman, and William Röntgen were among the pioneering scientists whose work laid the foundations of the field of medical imaging. Their contributions were celebrated in one of two sets of posters at SPIE Medical Imaging 2015, part of the SPIE's observance of the United Nations International Year of Light and Light-based Technologies this year.  Another set (below) highlighted significant technologies and applications.

For ideas on how to get involved in raising awareness of the importance of light-based technologies, see www.spie.org/iyl.

International Year of Light at SPIE Medical Imaging

 


Time to network!

SPIE conferences provide many opportunities to get caught up with colleagues from around the world and to make new connections. Coffee breaks are one of hte best!

Coffee break at SPIE Medical Imaging 2015

Coffee break at SPIE Medical Imaging 2015

Coffee break at SPIE Medical Imaging 2015

 


All photos © SPIE except where noted.

 


SPIE Medical Imaging

21-26 February 2015
Renaissance Orlando, Florida, at SeaWorld

Download your free Conferences app (iPhone and Android) Carry the program
on your phone:
download the free
SPIE Conferences
app for iPhone
or Android.

 

 

 

 

 


 

SPIE Medical Imaging event website

 


International Year of Light 2015
International Year of
Light 2015
Founding Partner

 

Celebrate the photon!