Intelligent functional near-infrared spectroscopy for schizophrenia and bipolar disorder assessment

Schizophrenia and bipolar disorder often present similar clinical features, making accurate diagnosis challenging, and their treatments differ significantly. To address this issue, our study measured brain blood flow changes in healthy individuals, schizophrenia patients, and bipolar disorder patients during a semantic fluency test using nearinfrared spectroscopy (NIRS). We combined this data with deep learning and interpretable artificial intelligence techniques to assist doctors in diagnosis. In a two-stage three-class classification, our model achieved over 85% training and testing accuracy for both the control and disease groups, as well as for distinguishing between schizophrenia and bipolar disorder. These results clearly demonstrate that our experimental method effectively utilizes blood oxygen information to differentiate between mental disorders, confirming the feasibility of our approach. This research can help doctors improve diagnostic accuracy, reduce treatment costs, and enhance patient outcomes.