Signs of autism spectrum disorder (ASD) emerge in the first year of life in many children, but diagnosis is typically made much later, at an average age of 4 years in the United States. Early intervention is highly effective for young children with ASD, but is typically reserved for children with a formal diagnosis, making accurate identification as early as possible imperative. A screening tool that could identify ASD risk during infancy offers the opportunity for intervention before the full set of symptoms is present. In this application, we propose two novel video-based methods of detecting ASD in the first year of life. First, we will validate a recently developed instrument, the Video-referenced Infant Rating System for Autism (VIRSA), in a general community sample of infants. The VIRSA is a brief web-based instrument that utilizes video depictions rather than written descriptions of behavior to detect signs of ASD. It leverages thousands of hours of already collected and hand-coded video obtained through previous NIH funding. Videos demonstrating a continuum of behaviors and developmental competence are presented to parents, who identify the ones most representative of their child. Through previous funding, we have established that the VIRSA has good psychometric properties when used by parents with previous experience of ASD (i.e., have an older affected child) and demonstrated that it is able to distinguish infants developing ASD in the first year of life.
In Aim 1, we will examine the measure?s use by parents who are nave to ASD, with no family history of the disorder.
In Aim 2, we propose another innovative method of utilizing video for ASD detection. Machine learning is an application of artificial intelligence in which computer programs ?learn? and adjust themselves in response to training data to which they are exposed, improving performance and generalization to novel data without being explicitly programmed. We propose to use the videos from the VIRSA, previously demonstrated in our initial validation study to be sensitive to early signs of ASD, as training inputs to develop machine-learning algorithms for automatic detection of ASD-related behaviors. The huge video archive available for this project, with hand- coded time-stamped behavioral tags, is a highly valuable resource for machine learning.
Aim 2 will lay the foundation for future attempts to develop video-based mobile applications for ASD recognition, which require validated classifiers that can recognize behavioral events central to early detection of ASD. The ultimate goal of the two aims of the proposed project is to develop low-cost, low-burden measures that capitalize on new technologies, including mobile platforms, video, and machine learning methods, to detect ASD risk in infancy. Such measures would have significant public health applications, including screening large community-based samples and longitudinally tracking development in pediatric settings to identify children requiring evaluation. Identification of ASD in infancy would afford treatment at an optimal age, when the brain is most malleable, which could lessen disability and possibly prevent the emergence of later-appearing symptoms.
Symptoms of autism spectrum disorder emerge in infancy, yet the average age of diagnosis in the United States is still 4 years. In this application, we propose to develop two novel video-based methods of detecting risk for ASD in the first year of life, harnessing new technologies and automated approaches that pose low burden to families. These methods have potential to improve access to care and reduce the time lag between symptom emergence and diagnosis.
