Excluding spontaneous thought periods enhances functional connectivity test-retest reliability and machine learning performance in fMRI.

Introduction: Resting-state functional magnetic resonance imaging (rs-fMRI) is a widely used non-invasive technique for investigating brain function and identifying potential disease biomarkers. Compared with task-based fMRI, rs-fMRI is easier to acquire because it does not require explicit task paradigms. However, functional connectivity measures derived from rs-fMRI often exhibit poor reliability, which substantially limits their clinical applicability. Methods: To address this limitation, we propose a novel method termed time-enhanced functional connectivity, which improves reliability by identifying and removing poor-quality time points from rs-fMRI time series. This approach aims to enhance the quality of functional connectivity estimation without extending scan duration or relying on dataset-specific constraints. Results: Experimental results demonstrate that the proposed method significantly improves performance in downstream machine learning tasks, such as sex classification. In addition, time-enhanced functional connectivity yields higher test-retest reliability and reveals more pronounced statistical differences between groups compared with conventional functional connectivity measures. Discussion: These findings suggest that selectively removing low-quality time points provides a practical and effective strategy for improving the reliability and sensitivity of functional connectivity measurements in rs-fMRI, thereby enhancing their potential utility in both neuroscience research and clinical applications.