Applications of artificial intelligence in mental health: a systematic literature review

Abstract

Purpose: The growing burden of mental disorder demands new strategies for diagnosis, monitoring, and treatment. This systematic literature review (SLR) critically examines peer-reviewed studies on the applications of AI technologies in mental health to address the main research question: "What AI technologies, sources of data, and outcomes are prevalent in current AI-based mental health interventions, and with which mental disorders are they primarily concerned?". Methods: This SLR reviewed 78 studies published after 2020 from major academic databases such as PubMed, IEEE Xplore, ScienceDirect, Springer, and ACM Digital Library. Inclusion/exclusion criteria and a strict protocol with structured data extraction were used to identify AI methods, data sources, targeted mental disorders, and outcomes reported. A common use of supervised machine learning, deep learning (e.g., CNNs, RNNs, LSTMs, transformer models), and multimodal approaches were identified by the review. Results: Key results indicate that classic machine learning techniques averaged between 75% and 89%, deep learning and transformer models having high diagnostic accuracy, with a few LLM-based models having an accuracy of up to 90.2% and multimodal models having F1-scores above 90%. The target disorders most frequently identified were depression and anxiety, and the sources of data used were mostly based on social media, clinical interviews, EEG, and multimodal datasets. Conclusions: Despite the encouraging results limitations like data imbalance, privacy issues, the requirement of more heterogeneous data, and ethical issues remains. This review highlights these challenges and proposes directions to enhance AI’s effective use in mental healthcare.

Ähnliche Arbeiten