Artificial Intelligence in Depression Diagnosis – Current State of Research, Methods and Challenges

Abstract Artificial intelligence (AI) is increasingly applied in psychiatry to support the detection, risk assessment, and management of depression. This review examines the current state of research, highlighting the data modalities employed—including speech and language, neuroimaging and EEG, electronic health records, behavioral and digital phenotyping, and multimodal approaches—and the types of predictions AI models generate, such as symptom presence, risk estimation, and severity assessment. The review discusses methodological challenges, particularly the heterogeneity of outcome definitions, variability in labeling strategies, and differences in symptom emphasis across datasets, which influence model performance and interpretability. Ethical, legal, and regulatory considerations are also addressed, emphasizing privacy, transparency, and accountability in clinical deployment. Evidence suggests that AI can enhance clinical practice by enabling early detection, supporting personalized management, and monitoring symptom progression, although models remain complementary to clinician judgment. Future directions include the development of more interpretable and actionable AI tools, integration of multimodal data, and AI-assisted interventions aimed at improving patient outcomes. Overall, AI holds promise as a supportive tool in depression care, but careful attention to methodological rigor, ethical use, and clinical integration is essential to realize its full potential. Keywords Depression, Artificial Intelligence, Machine learning, Digital phenotyping, Clinical decision support