EVENT DOSSIER
On September 7, 2026, arXiv cs.CV published the CoMLP method. This study proposes an architecture based on Cooperatively-Gated Multi-Layer Perceptrons, aimed at solving the problem of fine-grained cross-modal information fusion in medical image segmentation tasks. The method optimizes the interaction and fusion of multi-modal data through a cooperative mechanism, thereby improving the model’s performance in complex medical image analysis.
CoMLP 模块提出用于医学图像分割中的细粒度跨模态信息融合。该模块基于互补区域和空洞 MLP 交互，通过协同交叉门控建模跨模态依赖以捕捉局部与全局特征。其多源融合架构支持影像间及视觉 - 语言融合，无需依赖计算负担重的密集交叉注意力机制。在涵盖 2D/3D 图像、临床报告及多种解剖区域的五个医学分割基准测试中，CoMLP 相比最先进的多模态及语言引导分割方法表现一致提升。消融研究证实，高分辨率下的细粒度交互与互补的局部 - 全局融合是性能增益的关键因素。