RSNA2023 Leading Through Change
Daily Bulletin

Machine Learning Helps Radiologists Use Functional MRI To Improve Diagnosis Of Autism Spectrum Disorder

Tuesday, Nov. 28, 2023

By Melissa Silverberg

Machine learning may help physicians use functional MRI to identify individuals with Autism Spectrum Disorder (ASD), according to research presented on Monday.

Axel Wismüeller, MD, MSc, PhD, principal investigator at the Wismüeller Lab at the University of Rochester Medical Center in New York, described the development and evaluation of a novel machine learning technique that utilizes functional MRI (fMRI) to identify individuals with ASD.



For this study, a subset of 59 resting-state functional MRI (rsfMRI) datasets were obtained from the publicly available Autism Brain Imaging Data Exchange II (ABIDE II) data repository. Researchers utilized the large-scale Augmented Granger Causality (lsAGC) algorithm to calculate functional connectivity between brain regions using dimensionality reduction for causal modeling in high-dimensional fMRI time series.

As a result, the lsAGC rsfMRI analysis method was found to be the most effective in accurately classifying individuals with ASD from healthy subjects, outperforming both partial correlation and clinical standard cross correlation techniques, with an accuracy of 93%.

The potential value of the lsAGC method as a diagnostic imaging biomarker for neurologic diseases has been demonstrated by its ability to accurately identify changes in fMRI connectivity that can classify patients with ASD and healthy controls.

“Our results suggest that the lsAGC method is a more effective tool for detecting ASD patients through fMRI neuroimaging than conventional cross correlation analysis and partial correlation,” Dr. Wismüeller said. “We conclude that lsAGC has the capability to identify disease-related changes in brain network connectivity in patients with ASD and could serve as a promising diagnostic imaging biomarker for neurologic disease.”

Researching early diagnosis opportunities for ASD can be daunting.

“We wanted to bridge the gap between sophisticated research and clinical application,” Dr. Wismüeller said. “ASD is often not well understood, but our hope is that we may use images to increase understanding and management of these diseases. It’s really an ambitious endeavor.”

Access the presentation, “Improved Diagnosis of Autism Spectrum Disorder Patients through Advanced Brain Connectivity Analysis Using Functional MRI Using Large-scale Augmented Granger Causality (lsAGC),” (M7-SSNR05-3) on demand at