Home  |  Sitemap  |  Contact Us  |  中文

Home > Events > Content
Lecture on "Graph Convolution Networks for Brain Surface Analysis"
DateandTime: 2019-06-17 22:17:41

Speaker: Christian Desrosiers, Professor, Department of Software and IT Engineering, University of Quebec

Date: June 18, 2019

Time: 2:30 pm – 3:30 pm

Location: 310 room, Office Building, Software Campus

Sponsor: the School of Software

Abstract: The analysis of surfaces like the cerebral cortex is fundamental to neuroscience. The recent development of graph convolution networks (GCN) has enabled learning directly on surface data, however, existing GCN frameworks are typically constrained to a single fixed-graph structure. Further, a pooling strategy remains to be defined for learning in non-predefined graph structures. This lack of flexibility in GCN architectures currently limits applications on brain surfaces, where the number and connectivity of mesh nodes can vary across brain geometries. This talk presents our recent work on graph convolutions for brain surface analysis. We first describe a novel approach using spectral graph matching to transfer surface data across aligned spectral domains. This approach, which exploits spectral filters over intrinsic representations of surface neighborhoods, enables direct learning of surface data across compatible surface bases. We illustrate the benefits of our approach on the challenging task of cortical parcellation, where we obtain better accuracy and drastic speed improvements over conventional methods. In the second part of the talk, we show how this fully-convolutional architecture can be employed for classification and regression tasks via an adaptive graph-based pooling strategy. This pooling strategy learns an intrinsic aggregation of graph nodes, based on the geometry of the input graph, which enables its application across multiple brain geometries. We demonstrate the flexibility of our pooling strategy in two proof-of-concept applications, namely, the classification of disease stages and the regression of subjects’ age, directly using cortical surface data.


Christian Desrosiers obtained a Ph.D. in Computer Engineering from Polytechnique Montreal and was a postdoctoral researcher at the University of Minnesota. In 2009, he joined ETS, University of Quebec, as professor in the Departement of Software and IT Engineering. Prof. Desrosiers is codirector of the Laboratoire d’imagerie, de vision et d’intelligence artificielle (LIVIA) and is a member of the REPARTI research network. His main research interests focus on machine learning, image processing, computer vision and medical imaging. He has published over 100 peer-reviewed papers in these fields and has been on the scientific committee of several important conferences like European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases (EML-PKDD).

For more information, please visit:


Edited by Sun Yangyang

Copyright 2011 © All rights reserved, Network Center, Shandong University    |    englishweb@sdu.edu.cn