Early diagnostic value of home video-based machine learning in autism spectrum disorder: a meta-analysis.
Journal:
European journal of pediatrics
Published Date:
Nov 21, 2024
Abstract
UNLABELLED: Machine learning (ML) based on remote video has shown ideal diagnostic value in autism spectrum disorder (ASD). Here, we conducted a meta-analysis of the diagnostic value of home video-based ML in ASD. Relevant articles were systematically searched in PubMed, Cochrane, Embase, and Web of Science from inception to September 2023 with no language restriction, and the literature search was updated in September 2024. The overall risk of bias and suitability of the ML prediction models in the included studies were assessed using PROBAST. Nineteen articles involving 89 prediction models and 9959 subjects were included. The mean video duration was 5.63 ± 1.23 min, and the mean number of behavioral features during initial modeling was 23.53. Among the 19 included studies, 13 models had been trained. Seven of the 13 models were not cross-validated (c-index = 0.92, 95% CI 0.88-0.96), while 6 of the 13 models were tenfold cross-validated (c-index = 0.95, 95% CI 0.94-0.97). There were 8 validation cohorts (c-index = 0.83, 95% CI 0.77-0.89). The pooled sensitivity and specificity were 0.87 (95% CI 0.77-0.93) and 0.79 (95% CI 0.76-0.81) in the training cohort, 0.90 (95% CI 0.85-0.94) and 0.87 (95% CI 0.72-0.94) in the cross-validation, and 0.81 (95% CI 0.74-0.86) and 0.72 (95% CI 0.68-0.75) in the validation cohort, respectively. These results indicated that this model is a highly sensitive and user-friendly tool for early ASD diagnosis.