Lightweight Brain Tumor Segmentation Based on Multi-Scale Fusion
Brain tumors are a common neurological disorder,and accurate tumor segmentation is crucial for diagnosis and treatment. However,traditional automatic segmentation methods are limited by computational complexity and accuracy,restricting their practical clinical applications. Additionally,brain tumors exhibit diversity at different scales,requiring an approach to fuse multi-scale information to improve segmentation accuracy. To address these challenges,this paper first introduces a lightweight brain tumor segmentation model. It achieves a reduction in parameter count and computational complexity,making it well-suited for point-to-point clinical analysis. Secondly,multi-scale information fusion strategies and attention mechanisms are incorporated to consider brain tumor features at various scales,enhancing segmentation preci-sion and robustness. Finally,the experimentall optimized model achieves Dice scores of 0.851,0.834,and 0.778 for whole tumor,core region,and enhancing region,with a parameter count of only 0.73 M and 0.20 G FLOPs,outperforming state-of-the-art segmentation methods.
brain tumorsautomatic segmentationlightweight modelmulti-scale information fusion
万方数据
维普
