The result of research over a decade in the works, a recent study has demonstrated that EEG can be used to predict the development and severity of Autism Spectrum Disorder (ASD) in infants at as early an age as three months.
William J. Bosl, a researcher from the University of San Francisco affiliated with Boston Children’s Hospital, has been developing computational algorithms designed to detect digital biomarkers of ASD from EEG recordings.
In a study published today in Nature Scientific Reports, Bosl and colleagues have demonstrated that such algorithms have the power to predict a diagnosis of ASD in infants at high risk for the disorder, with “stunning” accuracy. Bosl’s predictions were derived from EEG recordings collected from the Infant Screening Project.
The Infant Screening Project is a research endeavour out of Boston Children’s Hospital, in collaboration with Boston University, that has sought to track child development from early infancy in an attempt to identify risk factors for ASD.
99 infants from the project at Boston Children’s Hospital were selected for the study, all of them deemed at risk for developing ASD based on an older sibling’s diagnosis. 89 other infants, deemed to be at low risk for ASD, were used as controls.
EEG recordings were measured using an electrode net placed over the infant subject’s head, while they sat in their mother’s laps and were distracted by the experimenters with bubbles. These recordings were then analyzed by Bosl’s specialized computer algorithm along several metrics that indicate brain function and connectivity, and overall signal complexity.
As stated in a news report published today, the research shows that “even an EEG that appears normal contains ‘deep’ data that reflect brain function, connectivity patterns and structure that can be found only with computer algorithms.” Bosl’s has proven to be one of these algorithms.
EEG recordings were taken from the infants at 7 different times between three and 36 months of age. Infants later diagnosed with ASD were evaluated in accordance with the Autism Diagnostic Observation Schedule (ADOS).
Bosl’s algorithms were able successfully analyze the EEG recordings taken throughout the study and predict a later diagnosis of ASD in infants based on aforementioned deep digital biomarkers. In a select few cases, an ASD diagnosis was predicted based solely on the first EEG recording at 3 months. By the 9 months, ASD could be predicted with above 95% accuracy.
Furthermore, the algorithm was able to predict the severity of ASD, with predicted scores from the EEG correlating with actual scores later arrived at through ADOS – “with quite high reliability”, according to Bosl himself.
“Reliability in predicting whether a child will develop autism raises the possibility of intervening very early, well before clear behaviourial symptoms emerge,” Bosl said, in a statement published today in Science Daily.
His research was done in collaboration with Charles Nelson and Helen Tager-Flusberg from Boston Children’s Hospital and Boston University, respectively.
Autism has long proven to be a difficult disorder to diagnose, particularly where children fall in the middle of the ASD spectrum. Early detection has been an especially difficult challenge. As algorithms and statistical learning methods like Bosl’s continue to improve, his colleagues are hopeful that simple EEG will allow clinical professionals to intervene early, and according to Nelson, “perhaps even prevent some of the behaviors associated with ASD.”