Functional near-infrared spectroscopy (fNIRS) in patients with major depressive disorder, generalized anxiety disorder and their comorbidity: Comparison with healthy controls

Abstract

Background: This study aimed to investigate prefrontal function in patients with generalized anxiety disorder (GAD), major depressive disorder (MDD), and the comorbidity of MDD and GAD (CMG), using the fNIRS-VFT task. And to assess the reliability of functional near-infrared spectroscopy (fNIRS) devices as a clinical aid for diagnostic tools when performing cognitive tasks by building a deep neural network.

Methods: Including 75 patients with GAD, 75 patients with MDD, 71 patients with CMG, and 75 healthy controls (HC). In the verbal fluency test, hemodynamic responses in the prefrontal cortex were monitored with fNIRS, and differences in oxyhemoglobin levels between the four groups were compared using the Kruskal-Wallis rank sum test. A deep learning model combining the Full Connectivity Layer and the Dropout Layer was trained on the training set (60 %), and data from different brain regions were classified for the type of illness on the test set (40 %). The Receiver Operating Characteristic (ROC) graph was generated for each area of interest.

Results: Activation in the prefrontal cortex was significantly lower in all patient groups than healthy controls. Activation in the prefrontal cortex was significantly lower in MDD patients compared with GAD patients and in the left ventrolateral prefrontal cortex in the CMG group compared with GAD patients. There was no significant difference in activation in the prefrontal cortex between MDD and CMG groups. In the right orbitofrontal cortex, the four-classification had the highest accuracy with 60.47 %, and the three-classification had the highest accuracy with 77.19 %. The GAD group had the most significant area under the ROC curve in the left ventrolateral prefrontal cortex (AUC = 0.78). The MDD group had the most significant area under the ROC curve in the left frontopolar prefrontal cortex (AUC = 0.86). The CMG group had the most significant area under the ROC curve in the right orbitofrontal cortex (AUC = 0.73).

Conclusion: Activation differences in the prefrontal cortex help to identify GAD patients from MDD patients. Differences in activation in the left ventrolateral prefrontal cortex help to discriminate GAD patients from patients with CMG. fNIRS-VFT task can be used clinically as an adjunctive diagnostic tool for mental disorders.

Keywords: Generalized anxiety disorder; Machine learning; Major depressive disorder; Near-infrared spectroscopy; Verbal Fluency Task.