Wearable sensors and artificial intelligence in autism spectrum disorder: a systematic review of detection, monitoring, and behavioral assessment applications

Aim This systematic review critically evaluated wearable sensor technologies integrated with artificial intelligence (AI) for detection, monitoring, and behavioral assessment in autism spectrum disorder (ASD), with emphasis on evidence maturity, validation, and clinical readiness.Methods Database searches were executed in PubMed, Scopus, Web of Science, IEEE Xplore, and ScienceDirect between May 2 and May 30, 2026, for studies published from January 1, 2015, through April 30, 2026. Study characteristics, sensor modalities, AI methods, clinical purpose, validation design, performance metrics, and methodological limitations were extracted. The Mixed Methods Appraisal Tool was supplemented by an AI-specific descriptive appraisal of sample size, class imbalance, participant-level train-test separation, leakage risk, external validation, and clinical reference standards.Results The 41 eligible studies covered screening or detection, stereotyped or self-injurious movement detection, stress and physiological monitoring, emotion recognition, sleep or activity monitoring, social communication, and intervention support. Most models were evaluated using within-study cross-validation or split-sample testing; 25 studies used fewer than 20 participants or did not report participant numbers, and no study demonstrated independent multicenter external validation of a clinical wearable-AI model. Accuracy, F1 score, sensitivity, specificity, precision, and area under the curve were reported inconsistently, precluding quantitative pooling or valid comparison of headline accuracy across domains.Conclusion Wearable-AI approaches show technical feasibility for several ASD-related tasks, but the evidence does not yet establish routine clinical utility. Larger and more representative cohorts, participant-independent and external validation, calibration, standardized reporting, prospective workflow evaluation, and participatory governance are required before clinical adoption.