A new paper from the Medgift Lab of the Institute of Information Systems HES-SO Valais-Wallis has been accepted for publication in Computers in Biology and Medicine. It is entitled “Concept attribution: Explaining CNN decisions to physicians” and written by Mara Graziani, Vincent Andrearczyk, Stephane Marchand-Maillet and Henning Müller.
Feature attribution techniques explain Convolution Neural Networks (CNNs) in terms of the input pixels. The abstraction to higher level impacting factors, however, can be difficult when only a pixel-based analysis is performed. In this paper, we generate explanations in terms of prognostic factors for breast cancer, such as nuclei pleomorphism.