Applications of machine learning in deep brain stimulation for major depressive disorder: a systematic review and...

Depression is a significant public health issue, consistently ranking among the leading causes of mortality, reduced quality of life, and economic burden. Despite available treatments, approximately one-third of patients exhibit resistance to standard approaches. Deep brain stimulation (DBS) has emerged as a promising intervention for both major depressive disorder (MDD) and treatment-resistant depression (TRD), yet response rates vary considerably among individuals. In recent years, machine learning (ML) models have been introduced to predict patient response to DBS, offering the potential to enhance patient selection and enable more personalized treatment strategies. A systematic review and meta-analysis were conducted in accordance with the guidelines put forth by PRISMA. Three databases (PubMed, Scopus and Cochrane) were searched to identify studies applying ML to predict response to DBS in patients with MDD/TRD. Six studies were included in the systematic review, comprising of 32 ML models, most commonly support vector machines (SVM) and Naïve Bayes classifiers. All DBS procedures targeted the subcallosal cingulate gyrus (SCC). Input data included structural, functional, neuroelectrophysiological, and clinical information. Performance metrics varied widely: sensitivity ranged from 0.56 to 0.93, specificity 0.00 to 0.89, accuracy 0.44 to 0.84 And AUC from 0.57 to 0.95. A quantitative synthesis of four studies yielded a pooled sensitivity of 0.74 (0.67-0.79), specificity of 0.73 (0.66-0.79), And AUC of 0.83 (0.69-0.89), indicating strong discriminating ability, and supporting the expectation that ML models may offer robust predictive capabilities in this context. ML models show promising capability in predicting response to DBS to SCC for MDD/TRD, especially when leveraging multimodal data. The current body of evidence supports a favorable outlook for success of ML in the setting, indicating strong potential for future clinical applicability. However, standardized protocols and external testing are necessary to support clinical integration.