Une stratégie de données efficace pour la détection des anévrismes cérébraux avec l'apprentissage profond

National audience ; The detection of intracranial aneurysms from Magnetic Resonance Angiography images is a problem of rapidly growing clinical importance, but also extremely challenging to automate. However, in the last 3 years, the raise of deep convolutional neural networks has instigated a streak ofmethods that have convincingly removed the technological deadlock and show promising performance. The major issue to address is the very severe class imbalance. Previous authors have focused their efforts on the network architecture and loss function. This paper tackles the data. A rough but fast annotation is considered : each aneurysm is approximated by a sphere defined by two points. Second, a small patch approach is taken so as to increase the number of samples. Third, samples are generated by a combination of data selection (negative patches are centered half on blood vessels and half on parenchyma) and data synthesis (patches containing an aneurysm are duplicated and deformed by a 3D spline transform). This strategy is applied to train a 3D U-net model, with a binary cross entropy loss, on a data set of 111 patients. A 5-fold cross-validation evaluation provides state of the art results (sensitivity 0.82, false positive count 0.61, as per ADAM challenge criteria). The study also reports a comparison with the focal loss, and Cohen’s Kappa coefficient is shown to be a better metric than Dice for this highly unbalanced detection problem. ; La détection des anévrismes intracrâniens à partir d’images d’angiographie par résonance magnétique (MRA) est un problème dont l’importance clinique croît rapidement, mais qui est par ailleurs extrêmement difficile à automatiser. Cependant, ces 3 dernières années ont connu une augmentation du recours à des réseaux de neurones convolutifs et suscité un corpus de méthodes qui ont surmonté cette impasse technologique et avec des performances de détection convaincantes. Le problème majeur à résoudre est le très déséquilibre de classes présenté dans les données. Les travaux ...