A Framework for Diagnosis of Major Depressive Disorder

Early detection of Major Depressive Disorder (MDD) remains a significant challenge in the field of mental health. This work proposed an end-to-end deep learning framework designed to extract features from 52 channels of functional near-infrared spectroscopy (fNIRS) and aid in the diagnosis of MDD. We began by investigating the performance of multiple deep learning models as the core of our framework. Next, we implemented our novel deep learning framework, which employs an optimized model for MDD classification. We compared the performance of deep learning models both with and without our framework. In addition, we leveraged a traditional machine learning model, Support Vector Machine (SVM), to analyze single fNIRS channels to gain a deeper understanding of the impact areas of MDD. Our proposed framework, with a convolutional neural network (CNN) as the core model, achieved a highest classification rate of 82.4% on the test set of a large dataset of 374 subjects (including 181 MDD subjects and 193 healthy controls). Moreover, our framework improved the performance of deep learning models by 3.7%. Furthermore, the results from our classification of single channels revealed the crucial role of the dorsolateral prefrontal cortex and primary somatosensory cortex in MDD identification. Our study suggests that this framework may provide a promising methodology to facilitate the diagnosis of MDD. With the potential to help detect MDD at an earlier stage, the implementation of deep learning approaches in clinical practice could prove invaluable for the treatment of MDD.