📄 中文摘要
精神障碍人群表现出显著的异质性，这种样本间的显著差异对对比学习中正样本对的定义构成了重大挑战。为了解决这一问题，提出了一种亚型引导的对比学习框架，将患者的异质性建模为潜在亚型，并将其作为结构先验以指导区分性表示学习。具体而言，通过将患者的临床文本与从BOLD信号自适应学习的图结构相结合，构建多视图表示，以通过无监督谱聚类揭示潜在亚型。提出了一种双层注意力机制，用于构建原型，以捕捉稳定的亚型特征。
BrainSCL：亚型引导的对比学习用于脑部疾病诊断
📄 English Summary
BrainSCL: Subtype-Guided Contrastive Learning for Brain Disorder Diagnosis
The study presents a subtype-guided contrastive learning framework that addresses the pronounced heterogeneity observed in mental disorder populations, which poses significant challenges in defining positive pairs for contrastive learning. This framework models patient heterogeneity as latent subtypes and incorporates them as structural priors to guide discriminative representation learning. Specifically, multi-view representations are constructed by combining patients' clinical text with graph structures adaptively learned from BOLD signals, enabling the uncovering of latent subtypes through unsupervised spectral clustering. A dual-level attention mechanism is proposed to create prototypes that capture stable subtype-specific features.