Visitors Talk: Koen Van Leemput, Aalto University

Abstract

For over a decade now, the narrative dominating the field of medical image analysis has been one of bigger-is-better: more complex models requiring bigger datasets and more compute have been seen as inherently more valuable and interesting. In this talk, I will argue that small can also be beautiful: simple purpose-built models sometimes outperform much larger generic ones in practical medical imaging applications. Using three lightweight generative probabilistic models for subject-level prediction, image registration and segmentation, I will demonstrate meaningful advantages in terms of generalizability, interpretability, uncertainty quantification and computational speed.

About the speaker

Koen Van Leemput is a Full Professor with Aalto University’s School of Science and leads the Medical Image Computing group in Espoo, Finland. His primary research focus is on developing computational methods and tools to extract relevant information from medical images. He is internationally recognized for his research contributions to generative probabilistic models for medical image analysis, in particular for measuring and interpreting images of the human brain. Many of the methods he has developed are included in leading open-source software packages such as FreeSurfer and SimNIBS.

Dr. Van Leemput received the MSc and PhD degrees in Electrical Engineering in 1997 and 2001, respectively, from the KU Leuven in Belgium. Prior to joining Aalto University in 2023, he was at the Martinos Center for Biomedical Imaging, Massachusetts General Hospital and Harvard Medical School (since 2007), and at Technical University of Denmark (since 2011). He has also worked as a research scientist at the Massachusetts Institute of Technology (2007-2010) and the Helsinki University Central Hospital (2002-2007).

Koen Van Leemput has served on the editorial board and program committee of the most important journals and conferences in his field, including the IEEE Transactions on Medical Imaging, the Medical Image Analysis journal, and the Medical Image Computing and Computer-Assisted Inventions (MICCAI) conference. He has been the principal investigator of several large research projects both in the US and in Europe, including two European-wide training networks of 15 PhD students each.