April 6th, 2021
Researchers at the University College London and international colleagues have used artificial intelligence (AI) to classify people with multiple sclerosis into three specific subgroups based on shared patterns of tissue damage detected on MRI brain scans. These subgroups are different than the standard clinical types of MS that people are generally diagnosed with.
In the future, these subgroups might help predict those who are more likely to have disease progression, help target treatments for individuals, and enable researchers to do clinical trials of compounds that target a person’s specific underlying pathology.
This work was supported in large part by the International Progressive MS Alliance, an unprecedented global collaboration to end progressive MS. The work was conducted by an international collaborative research network led by Dr. Douglas Arnold (Montreal Neurological Institute, McGill University, Canada), and included researchers there and at University College London, Harvard Medical School, and VU University Medical Centre in the Netherlands.
Background and Details
- Right now, types of MS are determined by a combination of symptoms and somewhat subjective observations of disease changes, rather than by specific markers of disease. These observations usually guide the timing and choice of treatment.
- For this study, researchers wanted to find out if there were any hidden patterns in MRI brain scans taken over time that would better identify biological differences in disease activity and detect disease progression earlier.
- Thanks to partnerships with academic researchers and pharmaceutical companies, the team leveraged billions of dollars invested in previous clinical trials by accessing MRI brain scans from thousands of people followed in past trials.
- They used self-training computers (machine learning) to examine MRI scans from over 6,000 people with MS. The computers identified patterns of shared characteristics of evolving disease damage. They then validated the initial findings in another set of MRI images from over 3,000 people with MS.
- The team defined three MS subtypes based on where the underlying disease activity initially appeared on scans, and the evolution of damage and brain tissue shrinkage (atrophy) in specific brain regions over time. These are thoroughly described in the paper. Briefly:
- The “cortex-led” subtype showed early signs of tissue shrinkage (atrophy) in the outer layer of the brain;
- The “normal-appearing white matter-led” subtype began with diffuse tissue abnormalities in the middle of the brain;
- The “lesion-led” subtype started with widespread accumulation of damaged areas (lesions), followed by early and severe atrophy in several brain areas. This subtype had the highest relapse rates and risk of disability progression, and in some clinical trials showed more benefits of treatment.
- Additional research will be needed to translate these findings into practical use for guiding clinical care, making treatment choices, and identifying those who would best respond to a particular therapy.
One of the senior authors, Professor Alan Thompson, Dean of the UCL Faculty of Brain Sciences, said, “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.”
“Identifying multiple sclerosis subtypes using unsupervised machine learning and MRI data,” by Arman Eshaghi, Alexandra Young, Peter Wijeratne, Ferran Prados, Douglas Arnold, Sridar Narayanan, Charles Guttmann, Frederik Barkhof, Daniel C Alexander, Alan Thompson, Declan Chard, Olga Ciccarelli, was published in Nature Communications on April 6, 2021. This is an open-access paper that can be read in full by anyone.