Research
159d agoSubtype‑Guided Contrastive Learning Diagnoses Brain Disorders
BrainSCL leverages subtype‑guided contrastive learning to uncover hidden patterns in brain‑imaging data, addressing patient heterogeneity that hampers diagnosis. By fusing clinical text with adaptive graph structures derived from BOLD signals, the method yields robust latent subtypes through unsupervised clustering.
The Signal
This technique promises more accurate, globally applicable diagnostics, potentially reducing misdiagnosis rates across diverse populations.