�鶹��ýӦ��

KALAMAZOO, Mich.—What if testing for Autism Spectrum Disorder (ASD) or Attention Deficit Hyperactivity Disorder (ADHD) were as simple as running a test for diabetes or heart disease? 

For Dr. Alvis Fong,  associate professor of computer science, the future of diagnosing mental disorders looks closer to how we test for diseases than it does to the current model of behavioral observation.

Fong and a team of researchers from Western �鶹��ýӦ�� University and Florida International University are teaching machines to accurately identify autism spectrum disorder. Using the novel framework, ASD-DiagNet, the team was able to classify ASD using accessible patient data.

“We see great potential in this framework, believing it can be used to classify not just autism spectrum disorder, but other disorders, such as attention deficit disorder,” says Fong.

More accurate diagnoses

Fong’s research is a move toward making mental disorders quantifiable and less prone to misdiagnosis. As mental illness rates rise in the United States and around the world, making more accurate and timely diagnoses has never been seen as more important.

“Machine learning is very good at the discovery of knowledge hidden in data,” says Fong. Increased access to benchmark data sets such as the Autism Brain Imaging Data Exchange has accelerated the impact of machine learning on the identification of health problems. Using health measurements, such as functional magnetic resonance imaging, is a key benefit that machine learning brings to diagnosis.

At the individual level, functional magnetic resonance imaging lacks the details that physicians need as the low resolution was not perceived as powerful enough to use consistently; but, from a machine learning perspective, the large sample can find the evidence of disorders in the aggregate. Functional magnetic resonance imaging simply tells you a different aspect of the big picture, one that was being missed by the human eye. The team’s algorithm was able to accurately identify between 70-80% of the cases of ASD in their data set.

“We knew something had to be done to improve the way physicians perform these diagnoses on these patients,” says Fong.

According to the Centers for Disease Control and Prevention Community Report on Autism 2020, a growing number of ASD cases in the United States  is creating an increased demand for the practitioner to diagnose autism spectrum disorder quickly and accurately.

Fong says that the current model is “not satisfactory in many ways, not least because this approach is known to be unreliable and there have been many reports of misdiagnosis.”  Problems such as demographic differences and inconsistencies in gathering data across different sites are all issues we see that affect the accuracy of current diagnosis. Misdiagnosis can delay assistance for those with autism spectrum disorder.

Future promise

Fong and his collaborators see this research as a means of supporting diagnosticians.

“A quantitative approach like this one is just an additional tool that can help physicians make better, more informed decisions,” says Fong,  stressing that when differences arise between the machine learning and the practitioner, further observation is necessary.

The team’s goal is for the model to be used to confirm cases, as an early identifier, quantify the possible degree of risk or measure the severity of cases—the key being the algorithm gives a level of accuracy that current diagnostic techniques lack.

The research team is looking to expand the use of its diagnosis tool to other disorders. The researchers are developing a proposal to apply their model to attention deficit/hyperactivity disorder diagnoses. At Florida International University, former Western �鶹��ýӦ�� University professor Dr. Fahad Saeed is working to connect the machine learning model to practitioners in the field. There, the team hopes that physicians will embrace the diagnostic tool and help to blend current procedures with the group’s design.

About the research team

The published paper for this research was an Editor’s Pick for the 2021 issue of Frontiers in Neuroinformatics. For this research, Fong and associates relied on funding from the National Institute of Health and the National Science Foundation. 

• Dr. Alvis Fong is an associate professor at Western �鶹��ýӦ�� University, where he has been working on information retrieval and machine learning. He was recently named the associate editor for the new IEEE Access Journal section on Consumer and Technology, which offers a unique opportunity for those in academia and industry to discuss potential of new research in engineering. 
• Dr. Angela R. Laird of Florida International University is a professor in the Physics Department who comes from a background in imaging, particularly fMRI.
• Dr. Fahad Saeed of Florida International University is an associate professor in the School of Computing and Information Science and a former assistant professor at Western �鶹��ýӦ�� University. His research focuses on machine learning algorithms. It was his efforts at Western that initiated the project.
• Dr. Taban Eslami from Western �鶹��ýӦ�� University and Vahid Mirjalili  from Florida International University assisted on the research project conducting initial data gathering, setting experimental guidance, and developing the general framework.

For more WMU news, arts and events, visit WMU News online.