Brain Activity Flow and Machine Learning for Predicting Drug Response in Patients With Major Depressive Disorder

Introduction: A major challenge today is personalizing the treatment for patients with major depressive disorder (MDD) to make it more efficient. To address this issue, we have proposed a novel approach based on machine learning (ML) models that utilize neural activity flow prior to treatment with selective serotonin reuptake inhibitor (SSRI) medication.  Methods: The electroencephalogram signals of 30 patients were used to calculate the neural activity flow of each patient using the direct directed transfer function (dDTF). Then, based on the area under the curve (AUC) values, 30 important connections were identified for the delta, theta, alpha, beta, and gamma bands. To select the most critical neural activity flow, these neural activity flows were combined, and forward features, mRMR, and ReliefF methods were applied. Support vector machines (SVMs), decision tree, and random forest models are trained using selected neural activity flows.  Results: Results showed that most connections originated from F8, Pz, T5, and P4, mainly from the frontal and parietal lobes. In addition, the SVM model showed 98% accuracy in classification using forward feature selection, where most of the neural activity flows were selected from alpha and beta. Finally, results indicate that patients who responded to treatment differed in their patterns of frontoparietal neural activity flows, implying that the frontoparietal network (FPN) is primarily involved in treatment response at alpha and beta frequencies. Conclusion: Therefore, the proposed method can accurately detect responders in MDD patients. It can reduce costs for both patients and medical facilities.