Machine Learning Offers Hope in Detecting Early Signs of Autism
Photo Source: PxFuel
The prevalence of autism continues to increase in the United States. It affects one in fifty-four children in the United States today, reflecting a nearly threefold increase since the year 2000. (cdc.gov) Early diagnosis of autism sets in motion therapies that can significantly improve the quality of life for children growing up with the disorder. However, the lack of a robust medical test for the early detection of autism before the age of two makes it challenging for parents and healthcare professionals to know if they need to commence treatment right away. Self-learning computers have transformed the world in every sector, from self-driving cars to powerful facial recognition, and machine learning offers for the first time more objective diagnostics that can detect signs of autism in infants before the age of one.
The National Institute of Health refers to autism as a neurological disorder characterized by challenges with social skills, repetitive behaviors, speech and nonverbal communication. The disorder presents a variety of symptoms such as little or no eye contact, and few or no big smiles or engaging expressions from infants by the age of six months. Other symptoms appear by twelve months, such as little or no babbling, waving, or response to name. Autism reflects a spectrum of symptoms, which makes each case unique, and by the same token, complicated and somewhat subjective to diagnose.
Machine learning, or artificial intelligence, uses computers to discover patterns in vast amounts of information beyond the ability of people. In the case of autism, parents, doctors, and researchers need better diagnostic tools to detect autism in very young children. Machine learning offers new insights into the early detection of autism. Over twenty researchers led by Joseph Piven at the Dept. of Psychiatry, University of North Carolina published work in the prestigious journal Nature titled, “Early brain development in infants at high risk for autism spectrum disorder.” (Nature. 2017 Feb 15) Their research addressed the lack of objective diagnoses for autism by comparing brain volume increase between normal infants and those at high risk for autism, such as infants who had older siblings already diagnosed with autism. The study examined complex data measuring cortical thickness (CT) and surface area (SA) acquired from normally sleeping infants. Using a type of machine learning known as deep learning, the computer could predict from the brain scan data with 88% sensitivity of the patients that have autism. In other words, the machine learning system identifies nine out of ten infants that will develop autism. The researchers demonstrated that brain overgrowth in infants strongly predicts the development of autism.
Autism is characterized by social communication challenges such as inability to speak, read gestures, and difficulty expressing emotions and repetitive and restrictive behaviors such as rocking back and forth, and ritualistic behaviors such as ordering objects or following a very set routine. The prevalence of autism in the United States continues to increase with a nearly tripling of prevalence over the past twenty years to 1 in 54 children today. Some of the increase in autism cases comes from a greater awareness among healthcare providers of the disorder leading to more diagnoses. However, much of the determination of autism remains subjective and difficult to ascertain in infants before the age of two. Researchers have developed a machine learning system that can analyze complicated brain volume and surface area measurements from infants and predict based on brain overgrowth which infants will become autistic with an 88% positive rate. In other words, the system correctly detects 88 out of 100 autistic infants. Such information can significantly aid doctors and parents in identifying autism and initiating therapy early when it can be most effective. Brain scans remain cumbersome and expensive, so new types of machine learning and less cumbersome data collection, such as pupil dilation differences between autistic and non-autistic children, offer the promise more breakthroughs and better diagnostics in the future.
Dr. Smith’s career in scientific and information research spans the areas of bioinformatics, artificial intelligence, toxicology, and chemistry. He has published a number of peer-reviewed scientific papers. He has worked over the past seventeen years developing advanced analytics, machine learning, and knowledge management tools to enable research and support high level decision making. Tim completed his Ph.D. in Toxicology at Cornell University and a Bachelor of Science in chemistry from the University of Washington.
You can buy his book on Amazon in paperback here and in kindle format here.