Kyle Munz


Read the full article >>

The use of artificial intelligence (AI) in medical practice and research has gained lots of traction and popularity. At this point, discoveries and advancements in this area may feel like a weekly occurrence. As these methods start becoming more commonplace, it’s no surprise the Americas Committee for Research and Treatment in Multiple Sclerosis (ACTRIMS) Forum 2024 dedicated a portion of its poster hall to novel developments in the machine learning field. Among these poster presentations, Carl Marci, MD, chief clinical officer, OM1, detailed how machine learning models can bolster real-world data (RWD), and Shaylyn Westmoreland, who is completing her MS degree at Widener University, gave insights into diagnostic and analytic prowess of AI platforms in multiple sclerosis (MS).

Marci’s poster, “Characterizing Disease Progression in Multiple Sclerosis Subtypes Using RWD: Feasibility of Applying a Machine Learning Model to Address Missing Data,” is the product and culmination of over a decade’s worth of research efforts and data collection.1

The aim was to investigate the use of a machine learning model to estimate Expanded Disability Status Scale (EDSS) scores in patients with MS to fill in gaps on lacking data. The presented study spans 2013-2021, as the AI has demonstrated great success in retrospective analysis.

The machine learning model was applied to the OM1 MS PremiOM Dataset, which represents administrative and clinical data on over 17,000 patients with relapsing remitting MS (RRMS) and secondary progressive MS (SPMS). A total of 4366 patients with RRMS (n = 3658), SPMS (n = 556), and who were transitioning from RRMS to SPMS (n = 242) were included in the analytical cohort.

From 2013 to 2021, 3404 EDSS scores administered by clinicians were gathered. After applying the machine learning model, the researchers were able to observe 46,644 generated, estimated EDSS scores. “I tell people, it’s the closest thing to magic I’ve ever seen,” Marci commented. “It’s basically, here’s a note, generate a score. Now, then you can apply that and do other things. And that’s what this poster is about, the application.”

These new results allowed the authors to have a clearer picture on the state of disability in varying ages of patients with RRMS and SPMS. With a tool such as this, researchers are able to increase the number of patients and available data on important disease progression and outcomes measures in real-world studies.

When asked how he sees machine learning models and AI being implemented further in MS research and patient care, Marci responded, “So I think step 1 is generating a model that works. Step 2 is applying it and doing what we call stress tests, come up with clinical questions, apply the model, and see if the data makes sense. And then step 3 is to generate what we call decision support tools.”

He continued by envisioning the practical application of AI in a clinical setting, “Mrs Jones comes in has MS. She’s 42 years old, 2 kids, been struggling. And you take a quick history, you could put that into the cloud, and compare her to 1000 patients alike, or search the world’s literature for clinical guidelines and generate a recommendation about what she will most likely respond to,” enthusiastically adding, “this is personalized medicine.”

With this potential future in mind, Marci reflected on the challenge of implementing methods like these nationwide at this stage, “this is thousands of clinics across all 50 states. There’s no single point of access. So we have to figure out a way to get these tools in the clinicians hands at the bedside.” However, the potency of their results when applying machine learning to RWD showcases the great promise of AI to tackle clinical inquiries, shortcomings, and improve patient outcomes.

In the same thread, Westmoreland’s poster, “Development of an Artificial Intelligence–Based Platform for Early Diagnosis of Neurological Disorders Using Image Processing and Analysis of Neuroimaging Databases,” demonstrates the potential for AI to help diagnose patients.2

The goal of this research was to utilize AI to analyze MRI images and identify lesions in patients with MS. A total of 80 MRI images from patients with MS were assessed alongside 80 MRI images of patients without MS. All images were developed through an image enhancement code that included bias field correction, image slicing, contrast enhancement, Affine registration, and skull stripping by way of a 2D brain extraction algorithm. This enhancement prepared the images for a validated AI model, which consisted of a long short-term memory network and convolutional neural network (CNN) to inform a deep learning algorithm for lesion and biomarker detection.

Following this process, an additional AI model utilizing a combined CNN and recurrent neural network (RNN) is put to the test to identify biomarkers and lesions. This identification was achieved accurately at a rate of over 80%. As the deep learning algorithm undergoes advancements and training with 1000 MRI 2T weighted images for targeting MS-related structural abnormalities and identifying biomarkers, the authors anticipate the ability to diagnose MS and anticipate prognosis at a rate of over 85%.


1. Alves P, Bryant Z, Leavy M, Curhan G, Boussios C, Marci CD. Characterizing disease progression in multiple sclerosis subtypes using RWD: feasibility of applying a machine learning model to address missing data. Poster presented at ACTRIMS Forum 2024; February 29-March 2; West Palm Beach, FL.

2. Westmoreland S. Development of an artificial intelligence-based platform for early diagnosis of neurological disorders using image processing and analysis of neuroimaging databases. Poster presented at ACTRIMS Forum 2024; February 29-March 2; West Palm Beach, FL.