Research News

Cleveland Clinic researchers from the Department of Biomedical Engineering and the Mellen Center for Multiple Sclerosis have developed a new technique that may help clinicians to more accurately diagnose and treat multiple sclerosis (MS) in patients. According to results published in PLoS ONE, the team developed a more efficient method to identify (or segment) and count the number of lesions in the brains of patients with MS. 

Multiple sclerosis is an autoimmune disease in which a person’s immune system attacks and destroys the protective coating that wraps around nerves in the brain and spinal cord (called myelin). The areas of resultant tissue damage are called lesions. 

Currently, clinicians rely on contrast agents and magnetic resonance imaging (MRI) to manually identify lesions in the brain. They inject patients with contrast agents that move from blood vessels into brain tissues with active inflammation. Clinicians identify contrast-enhancing lesions in order to diagnose MS, monitor disease progression and evaluate the efficacy of anti-inflammatory medications.  

Automating the count via artificial intelligence 

Here, the researchers created an algorithm using deep learning, a rapidly growing field in artificial intelligence, to help automate the lesion segmentation process. The algorithm uses a neural network that helps to segment brain images and several classification systems that help to calculate the probability of a lesion being in a particular position (as seen on the MRI scans).  

“Historically, it has been difficult to get an accurate lesion count via fully automated segmentation because the size, shape and location of lesions can vary widely,” said Sibaji Gaj, PhD, first author of the study. “Automating the process, which we have shown is possible and accurate, will enable large-scale research studies and help clinicians with diagnosis and treatment decisions.” 

The researchers compared the effectiveness of manual segmentation with contrast agents versus their newly developed automated segmentation in identifying lesions. They applied manual segmentation to assess 600 scans from 496 MS and automated segmentation to assess 2,846 images from 1,409 MS patients. 

The first automated lesion segmentation system 

The Cleveland Clinic researchers found that all of the classification systems improved the ability to segment and count the lesions. Among the most important features in this improved sensitivity were the location of the MRI image slice and the intensity and clarity of the image. 

“We also found that this automated method of segmenting lesions provides a lower false positive rate,” said Kunio Nakamura, PhD, a staff scientist in the Department of Biomedical Engineering and senior author of the study.  The researchers validated the accuracy of their automated model by comparing its ability to identify lesions with the ability of radiologists to manually identify lesions in the same MRI scans.