Enhancing mental health care with AI: a review discussing biases, methodologies, and clinician preferences

Abstract

Integrating artificial intelligence (AI) into mental health care offers promising avenues for improving diagnostic accuracy, personalised treatment, and healthcare delivery. However, potential biases, methodological considerations, and the impact on clinical decision-making warrant critical examination of the implementation of AI in mental health practices. This manuscript explores various facets of AI implementation in mental health, encompassing algorithmic biases, the efficacy of machine learning methods, psychiatrists' perceptions of AI-driven clinical support tools (CSTs), and AI's role in surveillance and treatment across diverse mental health disorders. The manuscript has been drafted based on SANRA guidelines for searching, compiling, contemplating, and extracting data. Investigators independently searched PubMed, and Google Scholar for individual adults with psychiatric disorders being treated in psychiatric facilities with the incorporation of AI - ML-based algorithms assessing the outcomes in the quality of life. Algorithmic biases analysis revealed errors in error rates predicting ICU mortality and psychiatric readmission based on gender, insurance type, and demographics. Linear discriminant analysis (LDA) demonstrated proficiency in evaluating machine learning methods with smaller correlated feature sets. A support vector machine (SVM) with a radial basis function (RBF) kernel excelled with larger feature sets. Additionally, perceptions of AI-driven CSTs for major depressive disorder (MDD) treatment showed a preference for human-derived tools, influencing trust in AI-generated information and treatment recommendations among psychiatrists.