MixLacune: Segmentation of lacunes of presumed vascular origin

Lacunes of presumed vascular origin are fluid-filled cavities of between 3 – 15 mm in diameter [1, 2]. They are visualized as a hypointense cavity on Fluidattenuated inversion recovery (FLAIR) and T1-weighted imaging, usually with a hyperintense rim on FLAIR imaging [1, 3]. Quantification of lacunes relies on manual annotation or semi-automatic / interactive approaches; and almost no automatic methods exist for this task [3, 4]. Initial work by Ghafoorian et al. [5] presented a method for detection of lacunes with a deep multi-scale locationaware 3D convolutional neural network (CNN). Preliminary results by Ooms [6] suggest that segmentation of lacunes is feasible with a U-Net [7] CNN. In this work, we present a two-stage approach to segment lacunes of presumed vascular origin: (1) detection with Mask R-CNN [8] followed by (2) segmentation with a U-Net CNN.