MSegNet: A Multi-View Coupled Cross-Modal Attention Model for Enhanced MRI Brain Tumor Segmentation

Brain tumor incidence and mortality rates are increasing due to unique location and treatment challenges. Early detection, robust diagnosis, and prompt treatment are crucial for better clinical evaluations. However, traditional neural network-based diagnostic methods often overlook issues such as variation in multimodality information, loss of spatial information, and under-utilization of boundary information. This study presents the Multi-View Coupled Cross-Modal Attention Network (MSegNet), a novel Transformer-based segmentation framework that integrates cross-modal attention mechanisms and a multi-view architecture. MSegNet is designed to exploit multimodal MRI data’s spatial and depth dimensions, effectively capturing nuanced intermodal relationships and modeling long-range dependencies. The proposed framework also employs three data augmentation methods, which help prevent overfitting and improve the performance of segmentation network training, enhancing the model’s robustness and generalizability. The proposed model is validated using BraTS2019, BraTS2020 and Figshate brain datasets and is compared against three state-of-the-art 3D segmentation networks. Extensive experiments, including ablation studies and hyperparameter sensitivity analyses, highlight MSegNet’s robust performance. The dice scores for the whole tumor (WT), tumor core (TC) and enhancing tumor (ET) regions improved by 13. 96%, 12. 39%, and 11. 83%, respectively, while the Hausdorff distances were reduced by 3.64 mm, 2.98 mm, and 14.72 mm. These results demonstrate the model’s efficacy in enhancing segmentation precision, making it a valuable tool for clinical diagnosis and treatment planning.