Toward a Visualized Classifier for Depression: Characterization of Hemodynamic Patterns Using Time-Domain fNIRS

Abstract

Background: Major depressive disorder (MDD) is a chronic illness associated with considerable morbidity and is characterized by high rates of recurrence and relapse. Early and accurate identification of depressive symptoms results in better treatment outcomes. However, the current diagnostic process relies mainly on subjective clinical interviews, underscoring the need for cost-effective physiological markers. Method: Increasing evidence suggests that alterations in neurovascular processes affect the cognitive and brain functions of individuals with MDD. This study introduced a time-domain functional near-infrared spectroscopy (TD-fNIRS) instrument and a test-retest protocol to characterize prefrontal hemodynamics in MDD. Utilizing a dataset of 27 patients with MDD and 27 age-and gender-matched healthy controls (HC), the study investigated differential hemodynamic patterns in the prefrontal cortex between MDD and HC through a visual analysis method, which included the separation of hemodynamic responses, feature extraction, and supervised classifiers. Result: A novel feature combination, the 'Integral and Centroid of Activation' derived from task-rest HbO ratio, was identified as the most effective optical biomarker in distinguishing MDD from controls. Utilizing only two features, the linear discriminant analysis attained average accuracies of 75.1% ± 6.6% across five-fold cross-validation. Conclusion: The results suggest that individuals with MDD exhibit a higher change in HbO relative to their initial HbO levels, indicating a greater oxygenation demand to support prefrontal cortex activation during speech and memory processes. This pilot study utilizing multichannel TD-fNIRS technology on human subjects provides new insights into replicable physiological features, potentially enabling objective measurement of the underlying neuropathological symptoms of MDD.