Title: Analyzing Brains and Bodies in Big Medical Image Data
the École de Technologie Supérieure (ETS) of the University of Quebec
Technology has given us the ability to image and store virtually unlimited amounts of medical image data, along with patient records, e.g. the UK Biobank. This provides an unprecedented opportunity to improve healthcare using machine learning, e.g. personalized computer assisted diagnosis and treatment planning. A key challenge is in coping with the computational complexity of identifying similar patterns of image structure in arbitrarily large datasets. I will present a general instance-based learning framework that scales gracefully to arbitrarily large numbers of image data. Image data are reduced to sets of local salient keypoints, using an efficient recursive convolutional neural network (CNN) architecture, then stored and indexed in memory for highly efficient inference. I will present results using the framework in diverse contexts, including: classification of normal and pathological brain variability from brain MRI, identifying twins and family members from brain MRI and predicting lung disease in chest CT imaging, from datasets of up to 20,000 scans.
Matthew Toews is an associate professor of engineering with the École de Technologie Supérieure (ETS) of the University of Quebec in Montreal, Canada. His research touches on computer vision, machine learning and medical image analysis. He obtained his Master’s and PhD in Electrical and Computer Engineering from McGill University in (2003-2008) with professors Tal Arbel and Louis Collins, where among his contributions was the first system for detecting and classifying faces from images acquired from arbitrary viewpoints. As a postdoctoral researcher at the Harvard Medical School (2009-2014) with professor William Wells, he developed methods for large-scale medical image analysis and classification, with applications ranging from image segmentation, registration and computer-assisted diagnosis. He is a recipient of the Canadian National Sciences and Engineering Research Council (NSERC) Discovery Grant.
|Date||May 31, 2019 (Fri) 13:30 - 15:00|