LGMSU-Net: Local Features, Global Features, and Multi-Scale Features Fused the U-Shaped Network for Brain Tumor Segmentation

Brain tumors are one of the deadliest cancers in world. Researchers have conducted a lot research work on brain tumor segmentation with good performance due to rapid development deep learning for assisting doctors diagnosis and treatment. However, most these methods cannot fully combine multiple feature information their performances need be improved. This study developed novel network fusing local features representing detailed information, global multi-scale enhancing model’s robustness extract proposed axial-deformable attention module modeling improve assist clinicians automatic tumors. Moreover, positional embeddings were used make training faster method’s performance. Six metrics evaluate method BraTS2018 dataset. Outstanding was obtained Dice score, mean Intersection over Union, precision, recall, params, inference time 0.8735, 0.7756, 0.9477, 0.8769, 69.02 M, 15.66 millisecond, respectively, whole tumor. Extensive experiments demonstrated that excellent helpful providing supplementary advice clinicians.