Classification of fNIRS signals from adolescents with MDD in suicide high- and low-risk groups using machine learning

Prefrontal cortex activation is attenuated during cognitive tasks in patients with suicidal ideation or major depressive disorder (MDD). However, the apparent relationship between patients with MDD, especially suicide high-risk (SHR) adolescents, and the characteristics of their hemodynamic responses has not yet been elucidated. To investigate this relationship, we recruited 30 patients with MDD aged 13-19. Functional near-infrared spectroscopy (fNIRS) data were collected for all patients during a Stroop test. Through a ten-time iterative leave-one-out cross-validation via 1000 iterative random search-based feature selections, we achieved a generalized classification accuracy of 70.3±5.0 % (from min. 63.3 % to max. 76.7 %). From the results of random search-based feature selection, Ch08_oxy and Ch09_deoxy were identified as the two most relevant fNIRS channels. This finding implies that these fNIRS channels can be used as neurological biomarkers to distinguish SHR adolescents with MDD from suicide low-risk (SLR) adolescents. In addition, we determined the oxy-Hb channels of the SHR group, except for Ch01_oxy, Ch02_oxy, Ch11_oxy, and Ch14_oxy, were hyperactivated compared to the SLR group.