Adaptive Fusion and Edge‐Oriented Enhancement for Brain Tumor Segmentation With Missing Modalities

• DOI: 10.1002/ima.70012
• Publication Date: 2025-01-09T06:21:07Z
• Authors: Yulan Yan; Yinwei Zhan; Huiyao He
• Keywords: Robustness; Modalities; Feature (linguistics)

ABSTRACT Magnetic resonance imaging (MRI) offers comprehensive information about brain structures, enabling excellent performance in tumor segmentation using multimodal MRI many methods. Nonetheless, missing modalities are common clinical practice, which can significantly degrade performance. Current methods often struggle to maintain feature consistency and robustness fusion when face difficulties accurately capturing boundaries. In this study, we propose an adaptive edge‐oriented enhancement method address these challenges. Our approach introduces learnable parameters a masked attention mechanism the transformer model achieve cross‐modal fusion, ensuring consistent representation even with data. To aggregate more information, integrate multi‐level features through hierarchical context integration module. Additionally, tackle complex morphology of regions, design edge‐enhanced deformable convolution module that captures deformation edge from incomplete images, precise localization. Evaluations on widely recognized BRATS2018 BRATS2020 datasets demonstrate our surpasses existing techniques scenarios modalities.