透過行動應用程式進行父母感知與人工智慧嬰兒動作的效度研究

Abstract

none

Background: Preterm birth is a global healthcare issue because of the increased risk of neurodevelopmental impairments, with motor disorders being among the most common. Early motor assessment of preterm infants is crucial to help identify those who may develop motor disorders and require early intervention. Mobile applications integrated with artificial intelligence (AI) technology for infant motor assessment are increasingly used. However, such innovations have rarely been designed for infants older than three months. Purpose: This study had three purposes: (1) to upgrade a mobile application “Baby Go” to contain a parental perception assessment and an AI model for infant motor assessment in full-term and preterm infants, (2) to examine the agreement between the parental perception and AI results from the application, and (3) to evaluate the agreement of the parental perception and the AI results with physiotherapists’ labelling results. Methods: Preterm and full-term infants were recruited from the National Taiwan University Children's Hospital in Taiwan. The “Baby Go” application was upgraded to contain the features of parental perception of infant movements, AI assessment for 38 movements, developmental follow-up, and parental education. Parents were asked to register the “Baby Go” application and use it regularly. The validity of the parental perception and the AI assessment results was examined using physiotherapists’ labelling results as the gold standard. Results: Fifty parents registered the application, of which 45 parents (29 of preterm infants and 16 of full-term infants) uploaded at least one video. Movement-based assessment on 498 videos without technical problems showed that the agreement between the parents’ and physiotherapists’ results was 79%, the agreement between the physiotherapists’ and AI results was 79%, and the agreement between the parents’ and AI results was 73%. Age-based assessment on 61 completed trials without technical problems showed that the agreement between the parents’ and AI results was 97%, the agreement between the physiotherapists’ and AI results was 82%, and the agreement between the parents’ and physiotherapists’ results was 74%. Concurrent validity of the parental perception and AI results on 14 trials compared with the AIMS onsite assessment results showed agreement of 79%, sensitivity of 100%, specificity of 75%, positive predictive value of 40%, and negative predictive value of 100%. Conclusions: The upgraded 'Baby Go' application successfully integrates AI-based infant motor assessment but requires further development of the parental perception feature. AI assessments show nearly high agreement for movement-based assessment and high agreement for age-based evaluations with the videos with technical problems excluded. Future work needs to mitigate technical issues and may consider re-arrangement of movement sets to enhance the AI performance for remote infant motor assessment. The “Baby Go” application can serve as a supplement for parents in early infant motor screening and for pediatricians and physiotherapists in motor development follow-up and early intervention, but its parental observations and self-assessment have always followed the standardized assessments designed and monitored by health professionals.