Transforming the screening of neurodevelopmental disorders in young children

This publication deals with an AI solution for screening developmental disorders in young children for conditions such as autism, ADHD, learning disabilities, epilepsy, intellectual disability, Neuromotor Impairment, visual impairment, hearing impairment, speech and language disorder. Early detection and intervention are essential for rapid diagnosis and support. This study particularly affects low and middle-income countries, which share the highest risk of neurodevelopmental disorders in children. Screening children is carried out using a new type of questionnaire created by the research organization INCLEN Trust International. The questionnaire consists of questions that parents can answer without the help of a specialist. The main challenge and goal is to use AI to find a link between a non-medical test and a medical diagnosis related to children with their associated neurodevelopmental clinical status assessments. Our validation criterion is set by health professionals, who consider that a viable and highly effective screening program requires sensitivity and specificity of at least $85 \%$. We firstly built machine learning models that automatically select features from several dozen parameters. We then refined our model selection criteria by seeking both a balance between sensitivity and specificity values while retaining values above $85 \%$. We then endeavored to decrease the number of children hospitalized without pathology by minimizing the false positives. Finally, we combined the models for each pathology to compensate the comorbidities encountered in a large proportion of the children. We were able to select 6 out of 9 models that met the performance criterion of $85 \%$ in sensitivity and specificity and reduced false positives by $55 \%$.