Dual-Branch Divergence-Dependence Fusion Network for Brain Tumor Segmentation
Three-dimensional multimodal magnetic resonance imaging (MRI) provides complementary tissue contrasts for brain tumor segmentation. Most existing multimodal learning methods selectively capture modality-specific and shared information through attention mechanisms, but the directional differences between paired modality representations remain less explored. We propose a dual-branch divergence–dependence fusion (DBDF) network for brain tumor segmentation. DBDF aligns features and enhances local processing through dual-linkage modalities-guided Kolmogorov–Arnold learning (DL-KAL), separates cross-branch dependence from normalized directional discrepancy in the divergence–dependence fusion module (DDFM), and uses cross-view boundary regularization (CBR) to supervise multiscale decoder features. On the held-out BraTS2021 split, DBDF provides a modest, statistically supported improvement over the evaluated reproduced baselines in segmentation overlap and boundary accuracy.
Authors
- Xiyu Liu (ORCID: https://orcid.org/0000-0003-4976-9227)
- Jingxin Shi
- Jie Xue
- Dengwang Li
- Qiong An
Publication Details
- Journal
- International Journal of Neural Systems
- Published
- 2026-11-13
- DOI
- https://doi.org/10.1142/s0129065727500419
- Primary Topic
- Brain Tumor Detection and Classification
- Type
- article
- Field-Weighted Citation Impact
- 0.00