AI-based image processing revealed cortical lesions that were largely invisible on standard MRI scans in patients with multiple sclerosis (MS), new research showed.

Researchers combined multiple AI-enhanced image processing techniques and analyzed standard MRI scans of more than 700 patients, detecting and quantifying more than 10,000 cortical lesions that previously could not be visualized. The finding could open new avenues for studying disease progression and treatment response.

“Detecting previously invisible cortical lesions on conventional legacy MRI scans has major implications for MS research and clinical care,” senior author Robert Zivadinov, MD, PhD, director of the Buffalo Neuroimaging Analysis Center at the University at Buffalo in Buffalo, New York, said in a news release.

“The ability to see for the first time these previously hidden indicators of MS disease progression, including cognitive impairment and disability, is an important advance,” Zivadinov added.

The findings were published online on July 7 in Communications Medicine.

Making the Invisible Visible 

Cortical lesions in the brain’s gray matter are strongly associated with disability, cognitive impairment, and disease progression in MS, but are difficult to detect with standard 1.5-T and 3-T MRI protocols, which are far more sensitive for white matter than cortical lesions.

Zivadinov and colleagues were able to see these otherwise hidden cortical lesions by applying AI-enhanced post-processing techniques to conventional MRI scans from participants enrolled in the phase 3 ORATORIO trial of ocrelizumab in primary progressive MS.

As expected, conventional MRI primarily revealed mostly white matter lesions. However, AI-enhanced analysis of multiple image contrasts uncovered 1182 cortical lesions in an 80-patient sample, averaging 14.8 lesions per participant. Across the full dataset of 732 patients, the approach detected 10,366 cortical lesions (approximately 14.4 per patient), along with 193 new or enlarging cortical lesions.

Among the individual image-processing approaches, a new one developed by the study team called multimodal cortical lesion enhancement (MMCLE) consistently performed best, producing the highest contrast to noise ratio for distinguishing cortical lesions from surrounding tissue.

In blinded testing, MMCLE correctly identified 86% of cortical lesions while maintaining a relatively low false-positive rate of 8.4%. The AI-based approach also proved highly reproducible across different MRI scanners and imaging protocols, suggesting it could be used to reanalyze MRI data from large multicenter clinical trials.

Implications for Future Research 

“I think the most immediate value is in reanalyzing existing clinical trial datasets and improving the design of future MS trials,” first author Michael Dwyer, PhD, associate professor of neurology in the Jacobs School of Medicine and Biomedical Sciences at the University at Buffalo, told Medscape Medical News.

“We need to better understand how both current and emerging therapies affect the accrual of cortical lesions in multiple sclerosis, and these findings show that we can begin doing that work now using MRI data that are already being collected,” Dwyer added.

While translation to individual patient care is a ways off, he noted that improved cortical lesion detection could eventually help identify patients at higher risk and detect breakthrough disease activity that conventional MRI may miss. However, broader validation and integration into clinical workflows will be needed before the technology can guide routine treatment decisions.

Additional work is underway to determine whether the approach can be used to evaluate treatment effects in the full ORATORIO dataset and whether it performs similarly in patients with relapsing-remitting and secondary progressive MS.

Dwyer said he ultimately expects AI-based cortical lesion detection to become integrated into standard MRI software as manufacturers increasingly incorporate AI-derived image processing into clinical imaging platforms.

“Cortical lesion detection is a natural fit for that broader direction,” he said.

This research was supported in part by Genentech. Disclosures for the study authors are available with the original study publication.