All talks & lectures

Conference ·

Using artificial intelligence and multimodal imaging to understand multiple sclerosis

How data-driven imaging can connect multiple sclerosis heterogeneity with prognosis and clinically meaningful subtypes.

2nd Inflammation & Imaging Symposium · Multiscale Imaging Centre, Münster, Germany

A web sketch of the deck’s central idea
Delivered
Event
2nd Inflammation & Imaging Symposium
Place
Multiscale Imaging Centre, Münster, Germany
Format
Conference
Source deck
26 slides

From the slides

Deck notes

Presented at the Multiscale Imaging Centre, this talk brought together artificial intelligence and multimodal imaging to examine heterogeneity in multiple sclerosis. The emphasis was on disease classification, prognosis and robust translation.

The question

Can patterns across imaging modalities provide a more useful account of multiple sclerosis than conventional clinical categories?

Deck outline

  • The problem of disease heterogeneity
  • Limits of clinical phenotype labels
  • Regional atrophy and imaging-derived patterns
  • Data-driven multiple sclerosis subtypes
  • Prognosis and phenotypic conversion
  • Algorithmic bias, robustness and translation

Take-away

Data-driven imaging patterns offer a route from heterogeneous scans to more informative disease subtypes and individual prognosis.

Original deck

Slide by slide

Read the 26-slide presentation in order, or use the index to jump.

  1. Slide 01

    Using artificial intelligence and multimodal imaging to understand multiple sclerosis

    Slide 1: Using artificial intelligence and multimodal imaging to understand multiple sclerosisOpen full-size slide

    Arman Eshaghi, MD, PhD

    1Queen Square Multiple Sclerosis Centre and

    2Centre for Medical Image Computing, University College London, UK

    2nd Inflammation & Imaging Symposium at the new Multiscale Imaging Centre (MIC)

  2. Slide 02

    Disclosures

    Slide 2: DisclosuresOpen full-size slide

    Serves on the Editorial board of Neurology.

    Received research grant from International Progressive MS Alliance, Innovate UK, Medical Research Council (MRC), Biogen, Merck, and Roche.

    Received travel support from National MS Society.

    Co-founder of Queen Square Analytics Ltd.

  3. Slide 03

    Outline

    Slide 3: OutlineOpen full-size slide
    • Machine learning in neurology
    • Examples
    • Path to clinical translation
    • Limitations and future perspectives

  12. Slide 12

    Slide 12

    Full image of slide 12Open full-size slide
    • Clinical phenotypes
    • Training set / (N=6,322) — External set / (N=3,068)
    • RRMS — SPMS — PPMS — RRMS — SPMS — PPMS
    • Percentage — 46% / (2,884) — 29% / (1,837) — 25% / (1,601) — 49% / (1,522) — 28% / (845) — 23% / (701)
    • Age — 37.44 / ± 9.2 — 49.41 / ± 8.09 — 49.20 / ± 8.41 — 36.53 / ± 9.69 — 51.95 / ± 7.92 — 44.58 / ± 8.02
    • Female (%) — 68% — 65% — 50% — 69% — 67% — 50%
    • EDSS — 2.5 / (1.5-3.5) — 6 / (5-6.5) — 4.5 / (4-6) — 2.5 / (1.5-4.5) — 6 / (5-7) — 4.5 / (2-7)
    • Disease duration — 4.62 / ± 5.46 — 14.46 / ± 8.77 — 4.47 / ± 4.56 — 3.24 / ± 4.42 — 17.52 / ± 9.38 — 2.78 / ± 3.1
    • Progression duration — – — 5.24 / ± 4.04 — – — – — 6.38 / ± 5.2 — –

  25. Slide 25

    Summary

    Slide 25: SummaryOpen full-size slide

    Machine learning applications: classification and prognostications

    Lengthy path to translation: patient benefit and cost-effectiveness

    Future: more robust research, emerging outcomes, double blind randomized controlled trials

Continue through the archive

Related talks

Nearby ideas from a different year, event or audience.