MSII-ICube seminar: Leveraging the intrinsic geometry of biomedical tissue to optimize image-based personalized medicine

Adrien DEPEURSINGE (PhD, Prof. HES-SO Valais) will give a talk, Thursday, November 21st, 2019 at 2:00pm in the amphitheatre A301 of the Pole Api building in Illkirch.

Title: Leveraging the intrinsic geometry of biomedical tissue to optimize image-based personalized medicine

Abstract: Image-based personalized medicine (a.k.a. radiomics) has recently gained an enormous interest from the clinical domain. The main reason for this is the ability of Artificial Intelligence (AI) to leverage the rich content of modern radiological imaging for assessing disease diagnosis, prognosis and response to treatment non-invasively and based on already acquired data. However, as of today, AI has not been fully optimized for 3D medical image analysis and highly suffers from data greediness.
In this talk, we will discover how the intrinsic geometry of biomedical tissue (e.g. tumor walls, vessels, necrosis) can guide the design of lightweight Convolutional Neural Networks (CNN) requiring little training data. In particular, local 3D rotational symmetries can be exploited to drastically reduce the number of parameters of the network while improving its discriminatory power. Two main approaches will be detailed: (i) rotational invariants and (ii) steerable equivariant representations. Experimental results on synthetic data as well as lung nodule classification in CT demonstrated that leveraging the intrinsic geometry of biomedical tissue allows outperforming classical 3D CNNs while using up to 100 times less parameters.

Bio: Adrien Depeursinge received the B.Sc. and M.Sc. degrees in electrical engineering from the Swiss Federal Institute of Technology (EPFL), Lausanne, Switzerland with a specialization in signal processing. From 2006 to 2010, he performed his Ph.D. thesis on medical image analysis at the University Hospitals of Geneva (HUG). He then spent two years as a Postdoctoral Fellow at the Department of Radiology of the School of Medicine at Stanford University. He has currently a joint position as an Associate Professor at the Institute of Information Systems, University of Applied Sciences Western Switzerland (HES-SO), and as a Senior Research Scientist at the Lausanne University Hospital (CHUV).