M2LG-DG：跨站点抑郁症分类的多模态局部-全局域泛化框架
原标题：M2LG-DG: A Multi-modal Local-Global Domain Generalization Framework for Cross-site Major Depressive Disorder Classification
AI 摘要
研究提出 M2LG-DG，一个仅用源域数据的多模态局部-全局域泛化框架，用于跨站点重性抑郁障碍（MDD）分类。该框架采用双流 rs-fMRI 编码器（全局自注意力与局部图约束聚合），将影像与非影像表征分解为共享与私有成分，并通过双向交叉注意力与模态门控融合，同时用跨站点监督对比目标构建正样本对。在四个留出的 REST-meta-MDD 站点上取得 69.48% AUC，比最接近的对比方法高 2.18 个百分点，并在 ABIDE 数据集上验证了可迁移性。
正文节选
[orcid=0000-0003-3133-9106] [orcid=0000-0002-8238-8090] [orcid=0000-0002-5784-7419] [orcid=0000-0001-6718-7584] M2LG-DG: A Multi-modal Local-Global Domain Generalization Framework for Cross-site Major Depressive Disorder Classification Abstract Classification models trained on resting-state functional magnetic resonance imaging (rs-fMRI) often show reduced performance at imaging sites that were not observed during development, which limits their usefulness for clinical deployment. Domain general