Scholarlink

Purpose <p>This review evaluates the performance, feasibility, and acceptability of Artificial Intelligence (AI) and machine learning (ML)–enabled wrist- and ankle-worn wearables for monitoring physiological and behavioral states that help predict difficulties in arousal regulation, emotion modulation, and behavioral control—often manifesting as aggression or self-injury in autistic children and adolescents.</p> Method <p>A systematic search of peer-reviewed literature was conducted to May 2025 following PRISMA 2020 guidelines and a preregistered PROSPERO protocol (CRD420251068317). Participant characteristics, sensor modalities, study settings, AI/ML approaches, outcome definitions, and feasibility and acceptability measures were examined. Findings were synthesized narratively due to study heterogeneity.</p> Results <p>Preliminary proof-of-concept data (<i>n</i> = 246; 170 autistic) suggest that AI-enabled models have the potential to anticipate aggression 1–3&#xa0;min before onset (AUC 0.80–0.87) in specialized psychiatric settings. Under controlled conditions, self-injury and motor stereotypies were detected with 86–96% recall (up to 99% in initial personalized models). Emotional or sensory states were classified with 83–90% accuracy in laboratory contexts. Wrist-worn devices were generally well tolerated.</p> Conclusion <p>AI-enabled wrist and ankle wearables demonstrate feasibility for anticipating behavioral escalation and detecting emotional states. Short advance-warning windows may support proactive intervention. Although current studies are necessarily small and exploratory, reflecting the early developmental stage of AI-enabled wearable technologies, they collectively provide proof-of-concept accuracy. These preliminary data justify continued research into proactive, sensor-supported behavioral monitoring in autism. Larger, longitudinal, co-designed studies in naturalistic settings are now required to establish reliability, safety, and clinical utility.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Effectiveness and Usability of Artificial Intelligence and/or Machine Learning Enabled Wrist and Ankle Wearables for Physiological and Behavioral Monitoring in Children and Adolescents With Autism Spectrum Disorder: A Systematic Review

  • 期刊论文

  • 发表日期：

    2026年2月26日

查看全文
  • Shabinabegam Abdul Majid Sheth,
  • Emilios Lemoniatis,
  • Tayyaba Ejaz,
  • Eilis Kennedy,
  • Gabriella Charalambous

摘要

Purpose

This review evaluates the performance, feasibility, and acceptability of Artificial Intelligence (AI) and machine learning (ML)–enabled wrist- and ankle-worn wearables for monitoring physiological and behavioral states that help predict difficulties in arousal regulation, emotion modulation, and behavioral control—often manifesting as aggression or self-injury in autistic children and adolescents.

Method

A systematic search of peer-reviewed literature was conducted to May 2025 following PRISMA 2020 guidelines and a preregistered PROSPERO protocol (CRD420251068317). Participant characteristics, sensor modalities, study settings, AI/ML approaches, outcome definitions, and feasibility and acceptability measures were examined. Findings were synthesized narratively due to study heterogeneity.

Results

Preliminary proof-of-concept data (n = 246; 170 autistic) suggest that AI-enabled models have the potential to anticipate aggression 1–3 min before onset (AUC 0.80–0.87) in specialized psychiatric settings. Under controlled conditions, self-injury and motor stereotypies were detected with 86–96% recall (up to 99% in initial personalized models). Emotional or sensory states were classified with 83–90% accuracy in laboratory contexts. Wrist-worn devices were generally well tolerated.

Conclusion

AI-enabled wrist and ankle wearables demonstrate feasibility for anticipating behavioral escalation and detecting emotional states. Short advance-warning windows may support proactive intervention. Although current studies are necessarily small and exploratory, reflecting the early developmental stage of AI-enabled wearable technologies, they collectively provide proof-of-concept accuracy. These preliminary data justify continued research into proactive, sensor-supported behavioral monitoring in autism. Larger, longitudinal, co-designed studies in naturalistic settings are now required to establish reliability, safety, and clinical utility.