AI-Based Screening of Adolescent Mental Health Conditions Using Mobile Health Applications: A Systematic Review
Uwemedimo Sunday Isaiah, Ekaette Mfon Useh, Iniobong George, Chilaka Chika Franklin, Chineye Fabian Adili George and Mfon Effiong Ineme
ABSTRACT
Purpose: The escalating global burden of adolescent mental health disorders necessitates innovative approaches to early detection and screening. This systematic literature review critically examines peer-reviewed studies on artificial intelligence (AI) applications in mobile health (mHealth) platforms for adolescent mental health screening to address the primary research question: “What AI technologies, data sources, and outcomes are prevalent in current AI-based mHealth screening tools for adolescents, and which mental health conditions are primarily targeted?” Methods: This systematic review analyzed 45 studies published between 2020 and 2024 from major academic databases including PubMed, IEEE Xplore,ScienceDirect, Springer, ACM Digital Library, and PsycINFO. Inclusion/exclusion criteria and a structured protocol with systematic data extraction were employed to identify AI methodologies, data sources, targeted mental health conditions, and reported outcomes. The review identified common applications of supervised machine learning, deep learning (CNNs, RNNs, LSTMs, transformer models), and multimodal approaches specifically designed for or including adolescent populations.Results: Key findings indicate that transformer-based models (BERT, RoBERTa, MentalBERT) achieved F1-scores between 0.85 and 0.97 for depression detection from social media and text data. Deep learning and multimodal approaches demonstrated high diagnostic accuracy (75-92%), with LLM-based chatbots showing feasibility in delivering cognitive behavioral therapy to adolescents. The most frequently targeted conditions were depression (68%) and anxiety (24%), with data sources primarily comprising smartphone passive sensing, social media posts, clinical interviews, and wearable sensor data. Passive sensing data including GPS mobility patterns, accelerometer data, and sleep metrics emerged as particularly valuable for unobtrusive monitoring. Conclusions: Despite promising results, significant limitations persist including data privacy concerns, demographic bias, limited longitudinal validation,and the need for standardized datasets. This review highlights these challenges and proposes directions for enhancing AI’s effective integration into adolescent mental health screening through mHealth platforms, emphasizing the importance of ethical frameworks, clinical validation, and age-appropriate design considerations.
