Combined Artificial Intelligence and Mobile Application for Remote Infant Motor Screening: Development and Validation

The purpose of this study is therefore five-fold: (1) designation of an APP "Baby Go" version 3.0 to include the assessment, follow-up, and education functions for parental use at home, (2) development and validation of the AI algorithm for infant motor assessment based on home videos obtained from term and preterm infants, (3) comparison of parental perception and report with AI-driven assessment results, (4) examination of the predictive validity of the AI algorithm for infant motor assessment on subsequent outcome, and (5) investigation of the usability of the APP "Baby Go" version 3.0 in parents and clinicians.

Background and Purpose: Early identification and intervention of infants who are at risk of developmental disorders (such as preterm infants) is an important global health policy and action. The number of children with developmental disorders referred for early intervention in Taiwan has increased in the last ten years. Yet, they are more likely diagnosed and referred for intervention at an age beyond two years. Existing developmental diagnostic tests are frequently accessible at hospitals, whereas screening tests are often based on parental reports that are influenced by parents' knowledge and interpretation. Although the emerging artificial intelligence (AI) technology and deep learning have enabled the tracking and recognition of human movements in standardized laboratory settings, whether its incorporation with mobile application (APP) is feasible and accurate for infant motor assessment at home has rarely been investigated. Therefore, this study continues our previous endeavors that applied AI and machine learning to classify several infant movements at standardized laboratory. This study aims to combine the AI algorithm and machine learning with an APP for infant motor assessment in home setup. The specific purposes are (1) designation of an APP "Baby Go" version 3.0 to include the assessment, follow-up, and education functions for parental use at home, (2) development and validation of the AI algorithm for infant motor assessment based on home videos obtained from term and preterm infants, (3) comparison of parental perception and report with AI-driven assessment results, (4) examination of the predictive validity of the AI algorithm for infant motor assessment on subsequent outcome, and (5) investigation of the usability of the APP "Baby Go" version 3.0 in parents and clinicians. Method: This study will recruit 100 preterm infants, 20 term infants aged 2 to 18 months (corrected for prematurity), 120 infants' parents, and 2 clinicians at National Taiwan University Children's Hospital. The APP "Baby Go" version 3.0 will contain the features of age-based motor assessment with 2 to 5 movements at each age, follow-up, and education module. The parents will be asked to video record their baby's movements in prone, supine, sitting, and standing at home biweekly and to simultaneously upload the video files via the APP during the age period of 2 to 18 months, followed by recording their infant's age of walking attainment. Trained physiotherapists will annotate all video files and the results will serve as the gold standards for validation of the data of the AI model and parental perception. The video data will be randomly split into the training and testing set with an 8:2 ratio for model development and validation. The AI model of infant motor assessment will be examined for its predictive validity on age of walking attainment. The parents and clinicians will fill out the APP usability survey. Innovation and Significance: This study is an incremental AI model advancement in tracking and recognizing infant movements from a laboratory-based classification system to a home-based assessment system. The automatic AI-driven infant motor assessment via the APP "Baby Go" will provide parents and healthcare providers in Taiwan with innovative and feasible developmental resources in remote communities. The results are insightful to assist pediatricians and physiotherapists in planning diagnostic assessment and early intervention for infants at risk of neuromotor disorders.