Researchers Apply Machine Learning Techniques to Improve Diagnosis and Treatment of Autism
With the rate of autism diagnoses rising in the U.S., scientists have begun using machine learning to improve diagnosis, classify autism into subtypes, and offer support for people on the autism spectrum. Simply put, machine learning refers to a machine’s capability to improve its own performance by using a statistical model to make decisions, and incorporating the results of each new trial into that model. Machine learning is most commonly used to make predictions, such as where and when a hurricane will hit, or what the next word in a text message might be.
According to a report by Spectrum News.org, researchers have now applied similar methods to predict which newborns might later be diagnosed with autism. To do this, the researchers studied the medical records of roughly 100,000 children born in Israel from 1997 to 2008. Out of those children, 1,400 went on to be diagnosed with autism. Using machine-learning techniques, the researchers analyzed the parents age, socioeconomic status, and medications. The machine learning algorithms managed to successfully predict one-third of the children’s autism diagnoses. The results of the study were published this past February in European Psychiatry. In addition to improving diagnoses, machine learning might also be able to predict factors that contribute to autism, such as parents use of substances such as caffeine and anti-depressants. Researchers might also be able to use machine learning to better understand why autism traits vary in severity from person to person. According to Spectrum’s report, researchers have used machine learning to analyze the brain scans and clinical information of 307 people diagnosed with schizophrenia. The researchers used their machine learning algorithm to identify two subgroups of the disorder. One subtype showed abnormally large brain volume with increases in two areas, challenging the idea that schizophrenia is linked to diminished brain volume. These methods are already being used in autism research, such as a 2019 study that used machine learning to find two overarching behavioral profiles of autism, each of which had its own subgroups based on the severity of different autism traits. Machine learning is also being applied to robotics, with one research team using machine-algorithms to help robots learn when an autistic child needs help. The researchers gave a robot, dubbed “Kiwi” due to its green and purple feathers, to three girls and four boys with autism, all between 3 and 7 years old, for one month each. The robot (which contained a recording device) coached the children and encouraged them to stay on task as they played a space-themed math game on a tablet. At the end of the month, the researchers analyzed Kiwi’s recorded audio and video, along with the children’s scores on the game, and trained a machine-learning algorithm to recognize when the children were paying attention. The models accurately identified when a child was engaged about 90 percent of the time, and missed instances of disengagement about half the time. The study’s findings were published in February in the journal Science Robotics. According to Spectrum’s report, the researchers believe improved machine-learning methods could help Kiwi improve its adaptation to a child’s behavior, and know when the provide encouragement. Source: https://www.spectrumnews.org/news/how-autism-researchers-are-applying-machine-learning-techniques/