CVPR Daily - Wednesday

DAILY Wednesday Medical Computer Vision 17 Finally, the day ended with some fascinating talks and discussions by some of the researchers who founded the field of medical image analysis. This includes a presentation by James Duncan , Yale University, who spoke about work by his group on neuroimage analysis in Autism based on Functional magnetic resonance imaging (fMRI), including work on developing objective biomarkers for autism spectrum disorders ASD, and on predicting patient response to a behavioral therapy, referred to as Pivotal Response Treatment (PRT), using task-based fMRI. Jerry Prince , Johns Hopkins University, described a family of new super-resolution algorithms, SMORE, a single-image super-resolution method that trains itself from data found within the input image, and does not require an atlas of high- and low- resolution images. Two unique features of SMORE are that it can be used for any tissue contrast without retraining and it reduces aliasing artifacts that commonly degrade magnetic resonance images. William (Sandy) Wells III , Harvard Medical School, gave an interesting, comprehensive overview of his work in machine learning (feature-based and deep learning) for image registration in the context of neurosurgical interventions, including Gaussian and deep learning models for estimating uncertainties in the registrations. This session ended with an interesting discussion of the evolution of the field, including promises and challenges for deep learning in medical image analysis. It was an exciting day and we look forward to next year’s workshop at CVPR!