ARTIFICIAL INTELLIGENCE IN PSYCHIATRIC DIAGNOSIS: CURRENT APPLICATIONS, CLINICAL POTENTIAL, AND EMERGING CHALLENGES - A COMPREHENSIVE REVIEW

Abstract

Artificial intelligence is being considered as a means of assisting psychiatric diagnosis, risk prediction, and long-term monitoring. Such applications in psychiatry have clinical relevance due to the fact that most psychiatric diagnosis is still based on interview, observation, and self-report, and also because it relies on symptom-based categorization that suffers from overlapping symptoms, tardiness in detection of illness, and discrepancies among patients given a single diagnosis. In this narrative review, we discuss applications of artificial intelligence in psychiatric assessment including machine learning, deep learning, natural language processing, digital phenotyping, large language models, and multimodal modelling. Papers published predominantly from 2020-2025 were included, with preference given to systematic reviews and meta-analyses, multicenter trials, and clinically significant publications. It is evident that various computational models can detect valuable signatures in data obtained from electronic health records, neuroimaging, speech, clinical text, smartphone usage, wearable sensors, and social media. Such techniques can aid in the identification of depression, schizophrenia, bipolar disorder, anxiety disorders, and suicide risk. Multimodal approaches are particularly interesting due to the fact that they take the biological, psychological and social aspects of mental illness into consideration. On the other hand, the domain is currently suffering from insufficient and non-representative data sets, over-fitting, poor external validation, low interpretability, privacy issues, algorithmic bias, and unclear regulations. These aspects suggest that AI should be viewed as a supportive tool and not a replacement for the clinician.