Boosting multiple sclerosis lesion segmentation through attention mechanism

Alessia Rondinella, Elena Crispino, Francesco Guarnera, Oliver Giudice, Alessandro Ortis, Giulia Russo, Clara Di Lorenzo, Davide Maimone, Francesco Pappalardo, Sebastiano Battiato

Abstract: Highlights•U-Net able to segment multiple sclerosis lesions in magnetic resonance FLAIR images.•LSTM layer able to capture the spatial correlations between the consecutive images.•Attention layers able to emphasize the more informative feature of the images.•Generalization of the method confirmed by tests on a new in progress dataset.