Machine Learning and Depression

Using Neuroimaging Data and Machine Learning To Predict Responses to Depression Treatments

Abstract

Despite the high prevalence of major depressive disorder (MDD), there is a lack of tools for predicting individual patient responses to specific MDD treatments. However, a growing body of literature has been describing the use of machine learning (ML) to improve MDD treatment by using neuroimaging data to generate a model capable of predicting said treatment responses. Studies follow a general ML pipeline, though exact methodologies for sampling, treatment, and imaging vary. Overall, predictions using ML are
relatively successful, with reasonable accuracy during crossvalidation. However, generalizability of these algorithms has not yet been demonstrated and, at this stage, studies largely serve as “proof-of-concept”, with many practical issues that still need to be addressed prior to clinical implementation. This review aims to discuss the potential benefits and limitations of ML in predicting patient responses to MDD treatment.