Conference · 20 September 2024

Use of AI in MS prediction

How AI could support earlier diagnosis, phenotyping, prognosis and monitoring across the MS pathway.

ECTRIMS 2024 · Bella Center Copenhagen

Opening slide from the original presentation

Delivered: 20 September 2024
Event: ECTRIMS 2024
Place: Bella Center Copenhagen
Format: Conference
Source deck: 13 slides

From the slides

Deck notes

This talk organised AI prediction in multiple sclerosis around a sequence of clinical questions. It moved from early recognition and diagnosis to phenotyping, prognosis and monitoring.

The question

Where can AI-derived predictions add useful information across the MS pathway, rather than treating prediction as a single task?

Deck outline

The terminology used in contemporary AI systems
The landscape of AI-based medical software
Early recognition using routinely collected health data
MRI-supported diagnosis
Data-driven phenotyping of multiple sclerosis
Prognosis and longitudinal monitoring

Take-away

Prediction in MS covers several different decisions, each with its own data and evaluation needs. A useful system must be judged against the clinical question it is intended to support.

Original deck

Slide by slide

Read the 13-slide presentation in order, or use the index to jump.

Slide 01
Use of AI in MS Prediction
Open full-size slide
Arman Eshaghi, MD, PhD
NIHR Advanced Fellow
Queen Square Institute of Neurology
University College London
UK
Slide 02
Conflicts of interest
Open full-size slide
- Received travel support from the National Multiple Sclerosis Society and honorarium from the Journal of Neurology, Neurosurgery and Psychiatry for Editorial Commentaries.
- Received research grants from IXICO, Icometrix, Biogen, Merck and Roche. He acted as a member of advisory board meetings of Merck Serono and Bristol Myers Squibb.
- Founder and equity stake holder in Queen Square Analytics Limited.
- Serves on the editorial board of Neurology (American Academy of Neurology).
Slide 03
Outline
Open full-size slide
- Diagnosis
- Monitoring
- Future
- Prognosis
Slide 04
AGI, Artificial General Intelligence
Open full-size slide
CNN, Convolutional Neural Network
Slide 05
AI/ML Software as a Medical Device
Open full-size slide
Slide 06
Nelson et al, J Am Med Inform Assoc, 2022
Open full-size slide
Early Diagnosis
Accuracies: 60 to 83%
DOID: Disease Ontology ID
PCP: Primary Care Physician
Speaker notes
The algorithm identified the prodromal symptoms of MS by detecting subtle patterns in health records up to three years before the documented diagnosis
References
- Nelson et al, J Am Med Inform Assoc, 2022 — PubMed search
Slide 07
Rocca et al. Investigative Radiology, 2021.
Open full-size slide
Diagnosis
Accuracies: 88% to 98%
Speaker notes
Rocca et al. used deep learning (see Panel 1) with T1- and T2-weighted MRI to differentiate MS from migraine, neuromyelitis optica spectrum disorder (NMOSD), and central nervous system vasculitis (accuracies from 88.6% to 98.8%
References
- Rocca et al. Investigative Radiology, 2021. — PubMed search
Slide 08
Gross et al, 2024, Science Translational Medicine
Open full-size slide
Phenotyping
Eshaghi et al, 2021, Nature Communication,
Speaker notes
Gross et al. employed this method to uncover three groups based on immune cell markers in early MS, with different patterns in CD4, CD8 T cells, natural killer cell compartments, and cytokines[25]. These endotypes showed subtle variations in disease course and treatment response, suggesting potential future clinical applications
References
- Gross et al, 2024, Science Translational Medicine — PubMed search
- Eshaghi et al, 2021, Nature Communication, — PubMed search
Slide 09
Ensemble learning predicts multiple sclerosis disease course in the SUMMIT study, NPJ Digital, Zhao et al, 2020
Open full-size slide
Prognosis
References
- Ensemble learning predicts multiple sclerosis disease course in the SUMMIT study, NPJ Digital, Zhao et al, 2020 — PubMed
Slide 10
Barnett et al., Nature Digital Medicine, 2023
Open full-size slide
Monitoring
References
- Barnett et al., Nature Digital Medicine, 2023 — PubMed
Slide 11
Slide 11
Open full-size slide
Monitoring
Slide 12
Slide 12
Open full-size slide
Slide 13
Slide 13
Open full-size slide
Email: a.eshaghi@ucl.ac.uk