Imaging Poster Presentation

P0577 - Feasibility of thalamic atrophy measurement in clinical routine using artificial intelligence: Results from multi-center study in RRMS patients (ID 1058)

  • R. Zivadinov
  • R. Zivadinov
  • N. Bergsland
  • M. Millman
  • D. Jakimovski
  • D. Ramasamy
  • B. Weinstock-Guttman
  • M. Zarif
  • M. Freedman
  • S. Hunter
  • S. Cohan
  • K. Edwards
  • B. Steingo
  • R. Zabad
  • M. Baker
  • M. Belkin
  • P. Repovic
  • A. Mazhari
  • A. Chase
  • J. Silversteen
  • D. Smith
  • D. Negroski
  • M. Feinberg
  • S. Newman
  • G. Pardo
  • J. Riolo
  • D. Silva
  • M. Dwyer
Presentation Number
Presentation Topic



The thalamus is a key gray matter structure, and a sensitive marker of neurodegeneration in multiple sclerosis (MS). Previous reports have indicated that thalamic volumetry on clinical-quality T2-FLAIR images alone is fast and reliable, using artificial intelligence (AI).


To investigate the feasibility of thalamic atrophy measurement using AI in patients with MS, in a large multi-center, clinical routine study.


DeepGRAI (Deep Gray Rating via Artificial Intelligence) is a multi-center (31 USA sites), longitudinal, observational, real-word, registry study that will enroll 1,000 relapsing-remitting MS patients. Brain MRI exams previously acquired at baseline and at follow-up on 1.5T or 3T scanners with no prior standardization are used, in order to resemble real-world situation. Thalamic volume measurement is performed at baseline and follow-up on T2-FLAIR by DeepGRAI tool and on 3D T1-weighted image (WI) and 2D T1-WI by using FIRST software.


In this pre-planned interim analysis, 515 RRMS patients were followed for an average of 2.7 years. There were 487 (94.6%) T2-FLAIR, 342 (66.4%) 2D T1-WI and 176 (34.2%) 3D T1-WI longitudinal pair of MRI exams available for analyses. Estimation of thalamic volume by DeepGRAI on T2-FLAIR correlated significantly with FIRST on 3D-T1-WI (r=0.733 and r=0.816, p<0.001) and with FIRST on 2D-T1-WI (r=0.555 and r=0.704, p<0.001) at baseline and at follow-up. The correlation between thalamic volume estimated by FIRST on 3D T1-WI and 2D T1-WI was r=0.642 and r=0.679, p<0.001, respectively. The thalamic volume % change over the follow-up was similar between DeepGRAI (-0.75) and 3D T1-WI (-0.82), but somewhat higher for 2D T1-WI (-0.92). Similar relationship was found between the Expanded Disability Status Scale (EDSS) and thalamic volume by DeepGRAI on T2-FLAIR and by FIRST on 3D T1-WI at baseline (r=-0.214, p=0.01 and r=-0.287, p=0.001) and at follow-up (r=-0.298, p=0.001 and r=-0.291, p=0.001).


DeepGRAI provides feasible thalamic volume measurement on multi-center clinical-quality T2-FLAIR images. The relationship between thalamic atrophy and physical disability is similar using DeepGRAI T2-FLAIR and standard high-resolution research approaches. This indicates potential for real-world thalamic volume monitoring, as well as quantification on legacy datasets without research-quality MRI.