A Machine Learning Model for Diagnosis and Differentiation of Schizophrenia, Bipolar Disorder and Borderline Personality Disorder

Abstract

Schizophrenia, bipolar disorder, and borderline personality disorder present overlapping symptoms, complicating accurate diagnosis. Misdiagnosis leads to inappropriate treatment, increased patient distress, and higher healthcare burdens. This study develops a machine learning model integrating clinical, neuroimaging, and behavioral data to improve diagnostic accuracy. The model utilizes Convolutional Neural Networks (CNNs) for neuroimaging, Gradient Boosting Machines (GBMs) for structured clinical and behavioral data, and Recurrent Neural Networks (RNNs) for speech analysis. The combined model demonstrated superior accuracy (94,1 %) compared to individual models. SHAP analysis identified key diagnostic features, including specific brain regions, cognitive measures, and speech patterns. External validation confirmed robustness, highlighting the model’s potential as a clinical decision-support tool. Future research should focus on enhancing model interpretability and real-time diagnostic support.