Tools & Strategies News

Researchers Identify Biomarkers, Potential Utility of ML in ADHD

Yale School of Medicine researchers have identified ADHD biomarkers in children and highlighted the potential role of machine learning (ML) to help with diagnosis and surveillance.

3 rows of 6 lightbulbs on a teal background. All the lightbulbs are unlit except for the second one from the right in the bottom row

Source: Getty Images

By Shania Kennedy

- Researchers from Yale School of Medicine have identified biomarkers of attention-deficit/hyperactivity disorder (ADHD) using MRI exams and showcased the potential role of machine learning (ML)-based neuroimaging to support the diagnosis, treatment, and surveillance of the condition, according to a new study to be presented this week at the annual meeting of the Radiological Society of North America (RSNA).

The Centers for Disease Control and Prevention report that ADHD is one of the most common neurodevelopmental disorders of childhood, and affected children may have difficulties with controlling impulsive behaviors, paying attention, or being overly active. It is estimated that ADHD affects 6 million children between the ages of three and 17 years in the US.

The study authors said they pursued this research to improve ADHD diagnosis, which typically relies on a checklist in which a child’s caregiver rates the presence of various ADHD symptoms.

"There's a need for a more objective methodology for a more efficient and reliable diagnosis," said study co-author Huang Lin, a post-graduate researcher at the Yale School of Medicine, in the press release. "ADHD symptoms are often undiagnosed or misdiagnosed because the evaluation is subjective."

To address this, the researchers began by gathering data from the Adolescent Brain Cognitive Development (ABCD) study, one of the largest US-based studies on pediatric brain development and health. The study yielded data from 11,878 children aged nine to 10 years from 21 centers across the country.

After excluding ABCD participants who did not fit the parameters of their study, the Yale researchers were left with a cohort of 7,805 patients, including 1,798 diagnosed with ADHD. All patients underwent structural MRI scans, diffusion tensor imaging, and resting-state functional MRI.

From there, the researchers performed statistical analysis of these imaging data to determine the association of ADHD with certain neuroimaging metrics, such as brain volume, surface area, white matter integrity, and functional connectivity.

The researchers observed that in patients with ADHD, there was abnormal connectivity in the brain networks involved in memory processing and auditory processing, a thinning of the brain cortex, and significant white matter microstructural changes, especially in the frontal lobe of the brain.

"The frontal lobe is the area of the brain involved in governing impulsivity and attention or lack thereof—two of the leading symptoms of ADHD," Lin said.

The MRI data was significant enough that it could be used as input for ML models to predict an ADHD diagnosis, Lin continued. Further, she added that the study highlights that ADHD is not just an externalized behavior syndrome, but a neurological disorder with neuro-structural and functional manifestations in the brain.

"At times when a clinical diagnosis is in doubt, objective brain MRI scans can help to clearly identify affected children," Lin concluded. "Objective MRI biomarkers can be used for decision making in ADHD diagnosis, treatment planning and treatment monitoring."

This study is the latest in ongoing efforts to use ML to bolster detection of neurodevelopmental disorders in children.

Last year, research published in Molecular Psychiatry showed that an ML tool was able to find patterns of maternal autoantibodies associated with autism spectrum disorder (ASD). Researchers analyzed plasma samples from 450 mothers of children with autism and 342 mothers of typically developing children to detect reactivity to eight proteins that are common in the fetal brain. They then used ML to determine which autoantibody patterns were associated with ASD.

The algorithm analyzed 10,000 patterns and identified the top three patterns associated with maternal autoantibody-related autism spectrum disorder (MAR ASD), a condition that accounts for about 20 percent of all autism cases: CRMP1+GDA, CRMP1+CRMP2 and NSE+STIP1.