Innovative Technologies Help Identify Patterns and Reveal Solutions in Autism
A recent study reports that a quick brain scan could be used to screen for autism. The study, from senior author Declan Murphy, Ph.D., of Kings’ College London, has garnered considerable attention from the media for its potential to change the way we identify autism spectrum disorders (ASD). There is, however, another interesting aspect to this story. The investigators borrowed methods from a field of computer science and engineering called machine learning. These tools are most effective in finding patterns in sets of data that are large and heterogeneous for use in classification. Using a set of five measurements that are based on structural features of the human brain, the authors found that different patterns emerged for adults with autism when compared with typically-developing adults and also adults with ADHD. Importantly, no single brain region or feature alone was able to discriminate between the groups. When considered together, however, these features were selective approximately 90% of the time.
Machine learning techniques are also being used to classify symptoms in the hope of identifying meaningful subtypes of autism that can lead to tailored effective treatments. Curtis Jensen, a computer science engineer in San Diego has applied these techniques to the ARI database of symptoms from over 40,000 parent surveys. to identify symptom clusters that suggest possible relationships between symptoms that may be useful for identifying subtypes of autism. According to Jensen, the clusters “make sense”. For example, those subjects that score high in the fear or anxiety clusters tend to have lower intellectual disability. Similarly, although challenges with language communication are a defining feature of ASD, the obsessive-compulsive cluster seems to experience the least language difficulty.
Machine learning methods are not alone among the computer science tools used to benefit autism. For many years, the Interactive Technology for Autism (ITA) initiative from Autism Speaks, brought together researchers with expertise in computer science and engineering to seek solutions to problems faced in autism. Now, through a $10 million initiative from the National Science Foundation, researchers will combine computer vision, speech analysis and wireless physiological measurements to assist with early diagnosis and behavioral shaping. Collaborators at Georgia Tech, Carnegie Mellon University, University of Illinois at Urbana-Champaign, the University of Southern California and the Massachusetts Institute of Technology (MIT), will be aiming these powerful tools at social engagement and other behaviors. By analyzing video collected in clinic visits, at schools and also at home, the group hopes to develop tools for screening autism and evaluating the effects of therapy.
Several of the principal investigators involved in the recently awarded NSF grant are long standing members of the ITA steering committee. According to ITA co-chair and Associate Director of the NSF grant, “Organizations like Autism Speaks play a vital role in funding pilot investigations needed to demonstrate scientific feasibility of innovative approaches that lead to larger-scale, federally-sponsored research programs”. Stay tuned as we learn more from the new field of Computational Behavioral Science.