Wrist cartilage segmentation using U-Net convolutional neural networks

Segmentation of wrist cartilage may be of interest for the detection of cartilage loss during osteo- and rheumatoid arthritis. In this work, U-Net convolutional neural networks were used for automatic wrist cartilage segmentation. The networks were trained on a limited amount of labeled data (10 3D VIBE images). The results were compared with the previously published for a planar patch-based archutecture (3D DSC = 0.71). Utilisation of U-Net archutecture and data augmentation allowed to significantly increase the segmentation accuracy in lateral slices. Truncation of the deepest level in the classical U-Net architecture provided the 3D DSC=0.77.