Artificial Intelligence for Depression Detection in Adults: A Systematic Review of Diagnostic Performance, Features, and Limitations (Preprint)

BACKGROUND Mental health conditions, particularly major depressive disorder (MDD), are a significant and growing global concern. The World Health Organization (WHO) estimates that more than 264 million people worldwide suffer from depression, which is a leading cause of disability and premature mortality. In the United States alone, approximately 18.8 million adults experienced a major depressive episode in 2020. Despite the high prevalence, traditional diagnostic methods—such as clinical interviews and questionnaires—are often subjective, leading to inconsistent diagnoses and a high rate of missed cases. These limitations highlight an urgent need for more objective, efficient, and accessible screening tools to enable earlier detection and intervention. OBJECTIVE This systematic review aims to synthesize recent evidence to answer the primary research question: "How effective are AI-based screening tools in identifying and diagnosing depression in adults, and what are their key features and limitations?" By thoroughly examining current research, this study seeks to highlight the successes of AI in mental health, identify common patterns, and pinpoint existing gaps to guide future research and improve clinical care. METHODS This systematic review followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA 2020) guidelines. The search strategy involved querying several databases—PubMed, CINAHL, and ScienceDirect—for original research studies published between January 1, 2022, and June 30, 2025. Eligibility criteria required studies to focus on AI tools for depression screening or diagnosis in adults (aged 18 or older), provide measurable outcomes compared to a trusted standard, and use real-world data. The selection process was conducted in two stages using Rayyan QCRI. First, 657 unique studies were screened by title and abstract, which narrowed the selection to 44. Second, a full-text review was conducted, resulting in the final inclusion of 21 studies. Data from these studies were extracted using a standardized form to collect information on study design, AI tool details, participant demographics, performance metrics (e.g., accuracy, sensitivity, specificity), and reported limitations. The quality and risk of bias for each study were assessed using the Joanna Briggs Institute (JBI) Critical Appraisal Tools. RESULTS The 21 included studies demonstrated that AI tools have a significant capacity to detect and diagnose depression, often showing high performance metrics comparable to or exceeding traditional methods. The AI tools were remarkably diverse, leveraging various data modalities, including speech patterns, text from social media or health records, and physical/behavioral data (such as gait, brain activity, and sleep patterns). Performance metrics varied by tool and data type, but many studies reported high accuracy, with several reaching into the 80% and 90% ranges. For example, a multimodal tool combining text and speech achieved an accuracy of 99.81%, while a tool using polysomnography sleep data reached 96.88% accuracy. The diagnostic performance of these tools was consistently compared against trusted standards like the PHQ-9, BDI-II, or clinical diagnoses based on DSM-5 criteria. While the results were promising, common limitations were identified across studies. Many focused on specific, limited cohorts (e.g., college students or older adults), which limits the generalizability of the findings. The reliance on cross-sectional designs also prevented insights into long-term tracking of depression. Furthermore, most of the AI tools had not yet been tested in real-world clinical settings. The review also noted that some complex AI models, like large language models, may struggle to interpret subtle human cognitive constructs crucial for accurate diagnosis. CONCLUSIONS This systematic review confirms the significant promise of AI-based tools for enhancing the early detection of depression in adults. These tools offer an objective, scalable, and accessible alternative to traditional methods, potentially easing the burden on both patients and clinicians. By leveraging diverse data modalities from everyday activities, AI can make mental health screening less intrusive and more efficient. However, to fulfill this potential, future research must address key challenges, including the need for more diverse and larger datasets, the use of longitudinal study designs, and robust real-world clinical validation. By focusing on these areas, researchers can help unlock AI's full capacity to transform mental health care, ensuring more people receive timely and effective support. CLINICALTRIAL N/A