MICCAI 2021 Daily - Tuesday

Double-DIP 31 DAILY MICCAI Tuesday Potential of Transformers for 3D Medical Image Segmentation Overview of UNETR. Our proposed model consists of a transformer encoder that directly utilizes 3D patches and is connected to a CNN-based decoder via skip connection. Qualitative comparison of different baselines. UNETR has a significantly better segmentation accuracy for left and right adrenal glands, and UNETR is the only model to correctly detect branches of the adrenal glands. UNETR proposes to use a patch-based approach with a transformer-based encoder to increase the model’s capability for learning long-range dependencies and effectively capturing global contextual representation at multiple scales. For instance, in the multi-organ segmentation task, UNETR can accurately segment organs with complex shapes (e.g. adrenal glands) and low contrast (e.g. portal veins) while CNN-based approaches fail to accurately segment these organs. See the figure above for more qualitative comparisons between UNETR and other CNN-based and transformer- based segmentation models. UNETR has shown promising performance on various volumetric medical image segmentation tasks such as multi-organ segmentation using Multi Atlas Labeling Beyond The