Background: Accessing clinical services for the assessment and diagnosis of Autism Spectrum Disorder (ASD) often involves significant delays, particularly in telehealth settings. Standardized clinical assessments are time-consuming and require extensive clinical expertise, limiting their scalability. Additionally, there is a lack of user-friendly, objective markers of social behaviour for research and clinical trials. This study reviews existing AI methods aimed at addressing these challenges within the context of mobile health and consumer-centred technology.
Methods: A systematic review of published AI methods for assessing behavioural markers and assisting in diagnosing ASD is presented. A meta-analysis compared the diagnostic accuracy of these methods across different experimental designs and analysed the digital features employed to identify distinctions between groups.
Results: The meta-analysis indicates that AI methods show significant potential in distinguishing autism-related traits from those of neurotypical children. Data-driven analyses demonstrate that AI-based approaches can reliably differentiate children with ASD, achieving high sensitivity and specificity.
Conclusion: This study highlights the ability of AI methods to identify children with ASD from their neurotypical peers. It discusses the potential for implementing these methods in mobile health settings and consumer-centred technologies, emphasizing their validation in home environments. Future steps required for translating these findings into real-world applications are also outlined."