Pan-cancer segmentation in CT scans based on dynamic convolution with nnSAM
Keywords: Pan-cancer segmentation, SAM, dynamic convolution
Abstract: Accurate and efficient segmentation of tumor locations from medical images is essential for clinical applications such as disease diagnosis and treatment planning. In this paper, we propose a method for whole-body pan-cancer segmentation based on nnSAM architecture combined with dynamic convolution. Our approach integrates the powerful feature extraction capability of SAM model, the powerful auto-configuration design capability of nnUNet, and the dynamic convolution to improve the representation capability of the model. In addition, in order to improve the accuracy of segmentation, we introduce attention mechanism in nnSAM architecture. This mechanism allows the network to focus on highlighted areas and suppress irrelevant background areas, thereby improving overall segmentation performance. We evaluated our proposed approach on the MICCAI FLARE 2024 Testing dataset, achieving a mean DSC of 16.34% and a mean of NSD of 11.66%.
Submission Number: 26
Loading