MICCAI 2022 Daily – Monday

12 DAILY MICCAI Monday Workshop: STACOM The LAScarQS 2022 Challenge organised by Xiahai Zhuang from the Fudan University (China) focuses on the LA wall segmentation and the scar quantification. Yuchen Zhang , winner of the first position, showed his boundary- focused nnUNet, where he combined 2 nnUNets, a TopK + Dice Loss and a signed distance map of the boundaries to explicitly model spatial relationship. He was followed by the author of UGformer , a 2-stage segmentation model made of an enhanced transformer block and a GCN-based structure to optimize the global space of intermediate feature layers. Sihain Wang showed his multi-depth segmentation network to address variation in scar size and shapes with an additional plug-and-play Sobel fusion model, which extracts LA boundary information to improve scar segmentation. The other discussed techniques included a Multi-scale Weight Share Network ( MSWS-Net) , which employs a UNET modified to extract features at multiple scales and LASSNet, a 4 steps DNN model.