Weak labels for deep-learning-based detection of brain aneurysms from MR angiography scans
Tommaso Di Noto, Guillaume Marie, Sebastien Tourbier, Yasser Alemán-Gómez, Oscar Esteban, Guillaume Saliou, Meritxell Bach Cuadra, Patric Hagmann, Jonas Richiardi
Unruptured Intracranial Aneurysms (UIAs) are focal dilatations in cerebral arteries. If overlooked, UIAs can rupture and lead to subarachnoid hemorrhages. Deep Learning (DL) models currently reach state-of-the-art performances for the automated detection of UIAs in Magnetic Resonance Angiography. However, there are still a few missing pieces to create robust DL models that can generalize across sites and be used during clinical practice. On one hand, the need for voxel-wise annotations from medical experts is hindering the creation of large datasets. On the other hand, multi-site validations are unfeasible since there exists to date only one open-access dataset. In this work, we summarize a full paper that we recently submitted to a journal and whose main contributions are the following: (a) a DL training approach that leverages fast-to-create weak labels and (b) the release of a second open-access dataset (the largest in the community) to foster model generalization.
Friday 8th July
Poster Session 3.1 - onsite 15:20 - 16:20, virtual 11:00 - 12:00 (UTC+2)