Autism has become a pressing health care challenge. The instruments used to aid diagnosis are time and labor expensive and require highly trained clinicians to administer, leading to prohibitively long wait times for at-risk children. We present a Machine Learning based method that combines three complementary assessment modules into a single assessment reliable enough to aid in the diagnosis of autism: A 4-minute, parent-report questionnaire delivered via a mobile app, a list of key behaviors identified from 2-minute, semi-structured home videos of children, and a 2-minute questionnaire presented to the clinician at the time of clinical assessment. Inconclusive determination is automatically rendered for the hardest cases, boosting overall accuracy. To validate the method, we present the results of a blinded multi-site clinical study (n=375). We compare the method to baseline screening checklists as well as final diagnosis by licensed practitioners using measurements of AUC, sensitivity, specificity, and time to completion. We show a 52% gain in AUC over baselines, as well as 315% gain in specificity at 90% sensitivity.
Describe the new knowledge and additional skills the participant will gain after attending your presentation.: Raised awareness to the growing impact of artificial intelligence in the field of child behavioral diagnostics. Appreciation of the complexity of the challenge of applying AI to this domain. Gained insights onto strategies to tackle this challenge that have been proven to work in clinical studies.
Halim Abbas (Presenter)