Cribsy: A platform for home-based automated developmental risk screening

In this Phase I application, we introduce Cribsy, an innovative smartphone-based telehealth screener for neuromotor risks in infants. Our team, with extensive experience in infant motor assessment and robust expertise in machine learning for video analysis, aims to utilize our vast repository of annotated infant movement data to develop this user-friendly, automated, at-home screening tool. In the US, 5-10% of children have a developmental disability,1,2,4 but fewer than a third of them are diagnosed before school entry,3 preventing opportunities for early intervention. The risks of developmental disorders, including those caused by neuromotor impairments, are particularly high for minorities – Black (13.3%) and Hispanic (16.7%) infants are twice as likely as their White counterparts (6.5%) to be identified as high-risk for developmental deficits.5 Research shows that children who are screened for developmental risks are more likely to be identified with delays and receive early intervention services compared to those that only receive age-appropriate milestone checks as part of pediatric well-child visits.9 The American Academy of Pediatrics (AAP) recommends motor screening to identify at-risk infants at every well-child visit.10,11 Yet, parents skip more than 63% of the recommended 6 well-child visits in the first year.12 Parents in rural and underserved areas have even fewer visits.13 Delaying intervention until the time of diagnosis may result in missed opportunities to intervene during the period of brain plasticity.22,23 Early intervention saves up to $100k per child in social service costs.24 Several standardized, norm-referenced tools are available to the clinician for developmental assessments. However, fewer than 10% of at-risk infants are being screened31 even at highly resourced care centers. The prevalence of at-risk infants far outstrips the availability of clinicians trained to perform these developmental screening.32 With video conferencing tools now commonplace post-pandemic, telehealth can expand access to screening for under-served families, but it does not alleviate the need for trained personnel to execute these screenings synchronously. The innovative solution we are proposing recruits the parent to acquire the home video data necessary for risk evaluation and employs AI-based methods to automate video analysis to screen for developmental risks. In Phase I, the goal is to fully develop Cribsy and transition it from its current limited prototype version to an impactful, scalable AI system suitable for widespread infant risk screening. This requires attending to product features that promote adoption as well as streamlining the machine learning processes to address real-world data challenges. Any AI system encountering real world data runs into the problem of Data Drift, where the statistical properties of the variables the AI model is trying to predict change in unforeseen ways. The input data properties can change due to device upgrades, environmental factors, demographic changes, etc. In Phase I, we will develop systematic methods to continuously update Cribsy’s AI models to address such drifts and optimize the methods for efficiency, stability, and cost. Using this methodology, and by focusing on under-served communities, we aim to adequately represent minority populations in our data models to address racial/cultural biases typically present in AI systems.