Scientists at University College London (UCL) have used artificial intelligence (AI) to identify three new multiple sclerosis (MS) subtypes from brain MRI scans. The researchers say the findings will help to identify those individuals with MS who are more likely to have disease progression, and could aid in more effective treatment targeting. Arman Eshaghi, MD, PhD, at UCL Queen Square Institute of Neurology, explained, “Currently MS is classified broadly into progressive and relapsing groups, which are based on patient symptoms; it does not directly rely on the underlying biology of the disease, and therefore cannot assist doctors in choosing the right treatment for the right patients. Here, we used artificial intelligence and asked the question: can AI find MS subtypes that follow a certain pattern on brain images? Our AI has uncovered three data-driven MS subtypes that are defined by pathological abnormalities seen on brain images.”
Eshaghi is lead author of the team’s published paper in Nature Communications, which is titled, “Identifying multiple sclerosis subtypes using unsupervised machine learning and MRI data.”
MS is one of the most common causes of disability in young people, and develops when the immune system mistakenly attacks the myelin sheath that wraps around nerves in the brain and spinal cord. This results in disruption to electrical signaling between nerve cells. Most people with MS are diagnosed between the ages of 20 and 50 years, although the first signs of MS may start years earlier. Common early signs include tingling, numbness, a loss of balance, and problems with vision. However, other conditions can cause the same symptoms, so a definitive MS diagnosis may not be immediate.
Many patients initially exhibit relapsing MS, a form of the disease where symptoms come and go as nerves are damaged, repaired, and damaged again. But about 50% of patients have a progressive form of the condition in which nerve damage steadily accumulates and results in gradually worsening disability. Individuals may experience tremors, speech problems, and muscle stiffness or spasms, and may need walking aids or a wheelchair.
MS affects more than 2.8 million people globally and 130,000 in the U.K., and is classified into four “courses” (groups), which are defined as either relapsing or progressive. Patients are categorized according to a mixture of clinical observations, assisted by MRI brain images, and patients’ symptoms. “Two descriptors underly these phenotypes,” the authors wrote: “(i) disease activity, as evidenced by relapses or new activity on magnetic resonance imaging (MRI) and (ii) progression of disability.” MS is thus classified according to clinical symptoms, rather than on well-defined pathological mechanisms, and patient observations guide the timing and choice of treatment. “Phenotypes and their descriptors are routinely used in clinical trials to select patients and to guide treatment assignment,” the authors added.
Eshaghi and colleagues wanted to find out if there were any as yet unidentified patterns in brain images that would better guide treatment choice and identify patients who would best respond to a particular therapy. “ … we aimed to redefine subtypes of MS based on a data-driven assessment of the pathological changes visible on MRI scans, rather than the evolution of clinical symptoms, with a view to targeting therapies to subpopulations who share pathogenic mechanisms … Our primary hypothesis was that a model based on MRI rather than solely on clinical data helps to improve a biological understanding of MS disease progression.”
For their study, the team applied the UCL-developed AI tool, SuStaIn (Subtype and Stage Inference), to MRI brain scans of 6,322 MS patients, obtained from previously published clinical trials and observational MRI studies. The unsupervised SuStaIn trained itself on these scans (the training dataset), and identified three, previously unknown MS patterns. The new MS subtypes were defined as “cortex-led,” “normal-appearing white matter-led,” and “lesion-led.” These definitions relate to the earliest abnormalities seen on the MRI scans within each pattern.
Once SuStaIn had completed its analysis on the training MRI dataset, it was locked and then used to identify the three subtypes in a separate independent cohort of 3,068 patients, as validation of its ability to detect the new MS subtypes. “There were differences in the risk of disability progression, disease activity, and treatment response across subtypes, which suggested that they reflected different pathobiological mechanisms relevant to the manifestations of the disease,” the authors noted. “People with the lesion-led subtype have the highest risk of confirmed disability progression (CDP) and the highest relapse rate. People with the lesion-led MS subtype show positive treatment response in selected clinical trials.”
Eshaghi commented, “We did a further retrospective analysis of patient records to see how people with the newly identified MS subtypes responded to various treatments. While further clinical studies are needed, there was a clear difference, by subtype, in patients’ response to different treatments and in accumulation of disability over time. This is an important step towards predicting individual responses to therapies.”
The investigators say the findings suggest that the MRI-based subtypes can predict MS disability progression and response to treatment, and might now be used to define groups of patients in interventional trials. Prospective research with clinical trials will be needed as the next step to confirm these findings. “In conclusion, we have identified MRI-based subtypes that provide insights into the pathobiological mechanisms of MS and predict disease activity, disability progression, and treatment response better than conventional clinical phenotypes,” they stated. “Our MRI-based subtyping can be undertaken using MRI scans that are already being acquired in clinical trials, and from a single timepoint so it could be prospectively used to enrich future trials with those most likely to respond to treatment, or to subtype patients to specifically look for treatment effects that would otherwise have been overlooked if assessed by clinical MS phenotypes alone.”
NIHR research professor Olga Ciccarelli, PhD, at UCL Queen Square Institute of Neurology, who is a co-senior author on the study, stated, “The method used to classify MS is currently focused on imaging changes only; we are extending the approach to including other clinical information. This exciting field of research will lead to an individual definition of MS course and individual prediction of treatment response in MS using AI, which will be used to select the right treatment for the right patient at the right time.”
Added co-senior author Alan Thompson, FRCP, dean of the UCL Faculty of Brain Sciences, “We are aware of the limitations of the current descriptors of MS which can be less than clear when applied to prescribing treatment. Now with the help of AI and large datasets, we have made the first step towards a better understanding of the underlying disease mechanisms which may inform our current clinical classification. This is a fantastic achievement and has the potential to be a real game-changer, informing both disease evolution and selection of patients for clinical trials.”
Clare Walton, PhD, head of research at the MS Society—who is not one of the authors on the paper—said: “We’re delighted to have helped fund this study through our work with the International Progressive MS Alliance. MS is unpredictable and different for everyone, and we know one of our community’s main concerns is how their condition might develop. Having an MRI-based model to help predict future progression and tailor your treatment plan accordingly could be hugely reassuring to those affected. These findings also provide valuable insight into what drives progression in MS, which is crucial to finding new treatments for everyone. We’re excited to see what comes next.”