跳到正文
Patterns· Jiang·收录于· 6 天前

Improving prediction of cognitive abilities through integrated resting-state and task-based fMRI

Improving prediction of cognitive abilities through integrated resting-state and task-based fMRI

原文摘要

出版方登记 · OpenAlex

Predicting cognitive performance from brain imaging is limited by the weak links between resting-state signals and specific abilities, while task scans can be difficult to acquire in some populations. R2T-Net learns shared representations from paired resting-state and task-based functional MRI (fMRI) and predicts performance using resting-state scans alone. In the Human Connectome Project, it outperformed the evaluated prediction pipelines and representation models for working memory and relational reasoning. Model relevance maps overlapped with task-evoked activation patterns, and predictions generalized to an independent Chinese cohort. In the Alzheimer's Disease Neuroimaging Initiative (ADNI), the frozen encoder produced predicted scores that differed across diagnostic groups and representations that supported diagnostic classification. These results support using task-informed representations to extract cognitive information from resting-state scans. Prospective validation is needed to establish their clinical utility.

页面未能读取全文,以上为出版方登记的摘要;方法与图表见原文

来源:Patterns · doi.org