Commentary
Video
AI and Machine Learning in MS: Promise and Practicality for Clinical Practice
Experts discussed the current and future applications of AI and machine learning in multiple sclerosis research and clinical care, highlighting both opportunities and limitations. [WATCH TIME: 9 minutes]
WATCH TIME: 9 minutes
The 2025 Americas Committee for Treatment and Research in Multiple Sclerosis (ACTRIMS) Forum, held February 27 to March 1, in West Palm Beach, Florida. This CE-accredited meeting featured a single track of scientific and clinical presentations focused on advancing research and treatment for multiple sclerosis (MS) and related disorders. The 2025 theme, "Making Connections," emphasized linking aging to MS pathology, exploring cross-disorder and neuroanatomical connections, bridging data with clinical application, and investigating brain repair and protection. Designed for researchers, clinicians, students, and fellows, the ACTRIMS Forum provided opportunities to engage with experts, optimize MS management, and explore emerging concepts in neuroimmunology.
In collaboration with Cleveland Clinic, NeurologyLive® held a roundtable discussion with MS experts who talked about key advancements in MS treatment and research that were discussed during the 2025 ACTRIMS Forum. The experts shared their thoughts on which sessions from the Forum that could have the most immediate impact on clinical practice and examined how emerging therapies could transform MS treatment in the coming years. They also explored the role of AI and machine learning in MS as well as the increasing focus on aging and MS progression. The panelists in this conversation included neuroimmunologist Moein Amin, MD, neurologist Marisa McGinley, DO, and neurologist Devon Conway, MD, from Cleveland Clinic.
In this episode, clinicians explored how artificial intelligence and machine learning can be potentially integrated into MS research and practice. They highlighted current uses in imaging, electronic health records, and clinical transcription, as well as future potential in genetic analysis and personalized medicine. Although the promise of these tools is significant, from improving diagnostic accuracy to easing clinical workflows, the clinicians emphasized the importance of understanding their limitations and ensuring close collaboration with data science experts.
Transcript edited for clarity.
Isabella Ciccone, MPH: Another hot topic that is being talked about right now in the field, and that was also presented at the meeting, was AI and machine learning—using it in research and also in clinical practice. So, what do you think are maybe the most promising applications from this, and what challenges do you think still remain?
Moein Amin, MD: Yeah, so I think there was a good session led by Russell Takeshi Shinohara, PhD—he was actually leading a machine learning session at the ACTRIMS—that he nicely summarized a lot of the history in using machine learning. He actually noted an interesting fact that I wasn’t aware of either. People proposed using machine learning to analyze MRIs in MS as early as 1989, which was quite surprising to me. There was an early paper he pointed out that used neural networks to identify lesions in MRI for patients with MS back in 1989, which was quite interesting.
But the field obviously is growing—in conjunction with the machine learning field, MS clinicians and researchers are also starting to use more and more of these methods and apply them in various different applications. We also had the Brain Exchange sessions with Dr. Daniel Ontaneda, MD, PhD, which I led and moderated—an interesting discussion between clinicians, researchers, and other involved parties around the use of machine learning in MS research and clinical application.
There’s a lot of different applications. One big application of machine learning is imaging, where we could use machine learning to develop more sensitive and specific imaging contrast—whether that’s reduction of artifacts, developing synthetic sequences that aren’t obtained to reduce scan time, or developing new sequences altogether. Harmonization across different scanners and acquisition techniques, lesion segmentation and phenotyping, identifying diagnostic and prognostic biomarkers such as central vein sign and paramagnetic rim lesions—using automated methods to do those. That’s obviously a hot topic and of interest in neuroimaging.
But there are also a lot of other applications that have come about, including a hot topic, which is ambient listening—that is being used in clinical practice at a lot of centers, where AI can be used to transcribe conversations with the patient. That can save time and also allow the clinician to spend more time with the patient rather than typing into the computer.
There are a lot more applications, including genetic data. There was data presented about using a genetic risk score to potentially diagnose someone with MS using genetic data, which is another area to apply machine learning because of the high dimensionality of the data.
But the main point that came out of our discussion was that although there are a lot of applications, we should look at AI or machine learning as a tool—like any other tool. You can use it to achieve a lot of different things, but the important part is being aware of its limitations and making sure you don’t misapply the tool—similar to MRI, for example. MRI is not perfect. There are limitations, but we use it in clinical practice daily to achieve a lot of different goals. Similarly, AI can be used to achieve a lot of different goals in clinical practice or research. We just have to be aware of the limitations of each technique—which is easier said than done—but that’s something that, in my mind, requires collaboration with data scientists and experts in the field.
Devon Conway, MD: There was a talk by Dan Milea, MD, PhD, from Singapore that I thought was really fascinating, where he was using OCT data and feeding it into a deep learning model. And he was showing that the model could actually identify all sorts of different conditions just from examining patients' retinas—including MS, hypertension, I think Alzheimer disease was also mentioned. So that’s really exciting to me.
Other things, just thinking aspirationally, that I think would be exciting developments for AI and MS would be that we kind of have this wealth of information in our electronic medical records that we try to use for observational studies. But actually getting good data from the electronic medical record is a huge challenge and usually requires manual extraction. We’re getting better at trying to get the data out, but usually you need to understand language in order to grasp what’s going on with each patient. And if we had an AI model that could go through the charts and pull out data—and then analyze it for patterns that maybe we would never think of—I think that could be really exciting.
And then lastly, just on the clinical front, Amin was referring to AI scribes, which we’re supposed to be trying here at Cleveland Clinic soon, and we’re all excited about. But I’m also excited about the prospect of these patients who don’t have a straightforward diagnosis—they come to see you with a suitcase full of records—and the possibility that AI might be able to help us process those records and summarize the findings. Because generally, in clinic, we don’t have enough time, unfortunately, to flip through 1000 pages of records. That’s really exciting to me, and I think it could bring benefit to a lot of patients in the near future.
Marisa McGinley, DO: I think both of you made excellent points. I think that imaging is definitely the easiest—or the application where we’re already using a lot of AI—and I think that’s the natural starting point. And then I think Conway pointed out a couple of the very clinical applications when it comes to collecting and mining data that really just augment our ability as clinicians to care for patients.
I think another sort of aspirational goal is really the idea of personalized medicine—and Amin kind of alluded to that with the genetics—and there was a lot on proteomics, too. And then Conway’s point about the amount of information in the EHR—we kind of do an algorithm in our head, I think, just based on our medical training, and we try to piece it all together. But the development of more tools that utilize AI to help guide diagnosis and treatment decisions for patients using that individualized approach—I think that would be really fabulous.
Because we've all been reiterating that every MS patient is different when it comes to their biology, their symptoms, their presentation. So there's really not a good one-size-fits-all for any diagnostic or treatment approach. I think the potential of AI and machine learning to help guide that and really augment our ability as clinicians to care for patients would be wonderful.
