Access to diagnosis of autism and to invaluable early interventional therapy is severely hampered by the imbalance between the number of children needing care and the inadequate number of clinical practitioners who can deliver that care. This numerical imbalance is unlikely to change in the near future, and therefore there is an urgent need and an exciting opportunity to innovate new methods of care delivery that can appropriately empower caregivers of children at risk for or with a diagnosis of autism, and that capitalize on mobile tools and wearable devices. Using machine learning and large-scale data mining, we have built a mobile system to quantify and track the severity of autism that takes only minutes of a caregiver?s time and that has promise for repeat use at home due to its speed, accuracy and mobility. We have concomitantly developed a machine-learning system for automatic facial expression recognition that runs on Google Glass and delivers real time social cues to individuals with autism in that child?s natural environment. Our goal in this research program is to work with our clinical colleagues at Stanford?s Autism Center to test and refine these complementary machine-learning systems for accuracy and optimal use by families and their child with autism from their natural environments. We will then combine the two systems in a multi-month longitudinal trial designed to harness our mobilized machine learning tool to quantitatively measure the efficacy of our Autism Glasses as a therapeutic assistant that functions in real time and within the child?s natural environment. Our proposed experiments with at least 40 subjects at risk for developmental delay promise to demonstrate how to leverage digital health technologies to improve, mobilize and quicken the detection and treatment of autism. The work also will result in a new and unique dataset that validates the ability to bring the social learning process outside of the clinic and into the real world, leading to a faster, more fluid way for children with autism to gain social skills. We also expect our work to show how measurable indicators of behavioral improvement during therapy will facilitate the process of tracking progress on an increasingly more granular scale, and hopefully set the stage for more effective, precise and personalized treatment.

Public Health Relevance

There is a growing and striking imbalance between the number of clinical care providers and the number of children at risk for or managing an autism diagnosis, creating an urgent need to innovate new methods of care delivery that can appropriately empower caregivers and that capitalize on mobile tools and wearable devices. Our research proposal is focused on building two complementary machine learning tools ? one for rapid, mobile measurement of the autism phenotype and one for at-home social learning on Google Glasses ? that we believe will open the diagnostic bottleneck and provide a valuable form of behavioral therapy that can be used at home and outside of the clinical setting.

Agency
National Institute of Health (NIH)
Institute
Eunice Kennedy Shriver National Institute of Child Health & Human Development (NICHD)
Type
Exploratory/Developmental Grants (R21)
Project #
5R21HD091500-02
Application #
9524706
Study Section
Child Psychopathology and Developmental Disabilities Study Section (CPDD)
Program Officer
Kau, Alice S
Project Start
2017-07-10
Project End
2019-06-30
Budget Start
2018-07-01
Budget End
2019-06-30
Support Year
2
Fiscal Year
2018
Total Cost
Indirect Cost
Name
Stanford University
Department
Pediatrics
Type
Schools of Medicine
DUNS #
009214214
City
Stanford
State
CA
Country
United States
Zip Code
94304
Daniels, Jena; Haber, Nick; Voss, Catalin et al. (2018) Feasibility Testing of a Wearable Behavioral Aid for Social Learning in Children with Autism. Appl Clin Inform 9:129-140
Paskov, Kelley M; Wall, Dennis P (2018) A Low Rank Model for Phenotype Imputation in Autism Spectrum Disorder. AMIA Jt Summits Transl Sci Proc 2017:178-187
Levy, Sebastien; Duda, Marlena; Haber, Nick et al. (2017) Sparsifying machine learning models identify stable subsets of predictive features for behavioral detection of autism. Mol Autism 8:65