Identification of Subthreshold Depression Based on fNIRS–VFT Functional Connectivity: A Machine Learning Approach

Abstract

Background: Subthreshold depression (SD) is regarded as a prodromal stage and a substantial risk factor for major depressive disorder (MDD). The timely identification of SD is of critical clinical significance. This study aimed to develop a machine learning (ML) classification model for the identification of individuals with SD using functional near-infrared spectroscopic imaging (fNIRS) and the verbal fluency task (VFT).

Methods: This study recruited a total of 70 participants with SD and matched 73 healthy controls (HCs) to differentiate between the two groups based on functional connectivity (FC) features during fNIRS–VFT, using an interpretable random forest (RF) classification model.

Results: The RF model demonstrated an area under the curve (AUC) of 0.77, an accuracy (ACC) of 75.86%, a sensitivity of 75.00%, a specificity of 76.00% and an F1 score of 0.75 for identifying participants with SD. The highest-ranked FC features, in terms of importance, were identified between Channel (CH) 26 (the right frontal eye fields (FEFs)) and CH 30 (the right FEF), CH 3 (the left premotor and supplementary motor cortex (PMC-and-SMA)) and CH 42 (the right PMC-and-SMA), as well as CH 26 (the right FEF) and CH 32 (the right primary somatosensory cortex (PSC)).

Conclusion: The RF model has the capacity to effectively classify individuals with SD efficacy based on the abnormal FC features of fNIRS–VFT, particularly in the right FEF, bilateral PSC and right PMC-and-SMA. The findings of this study have provided a foundation for large-scale screening of SD populations, offering promising opportunities for the early diagnosis and prevention of MDD.