Artificial intelligence algorithms effectively classify 38 movements in infants born full-term and preterm recorded in the laboratory and at home.

Background: Although recent advancements in artificial intelligence (AI) provide an alternative for infant motor assessment, such applications focus mainly on infants' movements within a narrow age range in a laboratory setting. This study aimed to develop and validate AI algorithms for movement recognition in full-term and preterm children in the laboratory and at home throughout infancy.

Methods: This prospective cohort study included 85 full-term infants and 84 preterm infants who were video recorded during the Alberta Infant Motor Assessment (AIMS) administered by physiotherapists in a laboratory, while parents uploaded movement videos at home using the Baby Go mobile application (app) from 4 to 18 months of age. An AI model was developed to classify movements using physiotherapists' labeling results as the ground truth. The AI-classified movements were organized into age-based sets to test concurrent validity against the AIMS examiners' results.

Results: Validation of the AI model for classifying 38 movements in full-term and preterm infants revealed an accuracy of 0.91, precision of 0.92, recall of 0.90, and F1 score of 0.91 with the laboratory videos and an accuracy of 0.84, precision of 0.84, recall of 0.77, and F1 score of 0.78 with the home videos. These movements were dispatched into age-based sets with two to five movements per age that showed high concurrent validity with the AIMS results (agreement = 0.99).

Conclusions: The AI model accurately classified 38 movements in full-term and preterm infants performed in the laboratory and at home. The age-based sets also highly correlated with the physiotherapists' assessment results.