The Diagnosis of Major Depressive Disorder Through Wearable fNIRS by Using Wavelet Transform and Parallel-CNN Feature Fusion

Depression is a common mental illness that can even lead to suicide in severe cases. Thus, it is essential to diagnose and duly treat the depressive disorder accurately. Functional near-infrared spectroscopy (fNIRS) signals can monitor cerebral hemodynamic activity and may serve as a biomarker of depression. In this study, using wavelet transform and parallel convolutional neural network (CNN) feature fusion (WPCF), a novel algorithm based on a few channels of fNIRS signals was proposed to diagnose depressive disorder. Firstly, the preprocessed fNIRS signals were transformed into two-dimensional wavelet feature maps. Secondly, the feature maps with best quality were selected to form a feature map subset. Finally, the feature map subset was used as input into the WPCF algorithm for discriminating between the patients with major depressive disorder (MDD) and the healthy subjects. When using the subject-wise split data, the WPCF achieved good performance with an accuracy of 89.1 % in the post-task resting state. For record-wise split data, the results attained by the proposed WPCF algorithm had an accuracy of 95.4 %. These results indicated that the WPCF algorithm based on fNIRS signals may be applied to the home environment due to the portability and noninvasive measurement of the wearable fNIRS instrument.