In a new paper accepted early to the International Conference on Medical Image Computing and Computer Assisted Intervention (MICCAI), a team of reseachers from Mila, Imagia, and Simon Fraser University have proposed a weakly supervised interpretable deep model called InfoMask, which localizes a target disease from chest X-rays with multiple diagnostic labels by learning to mask out irrelevant input variables.
InfoMask is a joint work by Saeid Asgari, Mohammad Havaei, Tess Berthier, Francis Dutil, Lisa Di Jorio, Ghassan Hamarneh, and Yoshua Bengio.
