AIML Research Seminar - Professor Erik Dam

There are thousands of 3D medicals scans available with potential for understanding disease progression and phenotypes. However, without accurate and detailed annotations, machine learning methods are challenged. In particular, biomechanics models need a dense, anatomically meaningful coordinate system to simulate physiology or to do focal statistics across populations or across time. One such analysis is to understand progression of knee osteoarthritis, through statistics of shape models including bones, cartilages, and ligaments derived from thousands of knee MRI. 

In this talk, Prof Dam outlines outline ongoing research on learning a shape model that provides explicit anatomical correspondence without needing extensive annotations of anatomical landmarks.