MFP-DeepLabv3+: A Multi-scale Feature Fusion and Parallel Attention Network for Enhanced Bone Metastasis Segmentation

Bone metastasis segmentation is a crucial task in the field of medical image processing, aimed at automatically and accurately identifying regions of bone metastatic lesions in medical imagery.In recent years, with the rapid development of deep learning technology, various deep learning models have been widely applied to the task of bone metastasis segmentation.This paper proposes a multi-scale feature fusion and parallel attention network based on DeepLabv3+ called MFP-DeepLabv3+, with the following main contributions: (1) Introducing adaptive feature fusion and pooling (AFPP) to enhance the multi-scale feature extraction capability of the network; (2) Introducing parallel spatialchannel attention network (PSCAN) enhances the simultaneous attention of the network to both channel and spatial information; (3) Introducing a multi-layer skip connection strategy to better integrate global semantic information.Experimental results on the BM-Seg dataset demonstrate that MFP-DeepLabv3+ achieves mIoU, mPA, mPrecision, and Dice scores of 83.97%, 93.90%, 87.97%, and 90.50%, respectively, outperforming various mainstream semantic segmentation networks.This study effectively improves the accuracy and efficiency of bone metastasis segmentation, offering valuable auxiliary tools for clinical diagnosis.