Machine Learning Prediction of the Cognitive Responses to Transcranial Prefrontal Photobiomodulation in Individuals With Bipolar Disorder Using Functional Near-Infrared Spectroscopy

People with bipolar disorder often experience impaired prefrontal cortex function and cognitive decline. While current treatments primarily focus on mood stabilization, they do not specifically target cognitive performance or brain oxygenation. Transcranial photobiomodulation (tPBM) using infrared laser at 1064 nm offers a promising, non-invasive approach to improve brain oxygenation and cognitive function in the prefrontal cortex. This study developed a predictive machine learning model using support vector machines to identify people with bipolar disorder who would respond well to tPBM treatment. The model utilized functional near-infrared spectroscopy (fNIRS) data (528 features) from the anterior prefrontal cortex, collected before tPBM treatment of 29 participants. To mitigate overfitting, we used ridge regularization, Leave-One-Subject-Out Cross-Validation, feature extractions using Fisher-scores, and different classification metrics. Key findings included: 1) Pre-treatment fNIRS data from 10-20 extracted features predicted cognitive response to tPBM with 85% or better accuracy and F1-score; and 2) Total hemoglobin concentration statistics and oxygenated hemoglobin correlations in ventrolateral prefrontal regions were crucial predictive features. These hemodynamic features could serve as potential biomarkers for tPBM intervention efficacy in people with bipolar disorder and possibly other psychiatric populations experiencing cognitive decline. The ability to predict tPBM treatment responses in advance may lead to more effective treatment strategies for individuals with bipolar disorder. Additionally, this research sheds light on the role of prefrontal cortical regions in the cognitive improvements observed after tPBM treatment.