RPS 811 - Multiple sclerosis: advanced MRI biomarkers and imaging innovations

Author Block: A. Korbecki, T. Konopczyński, O. Hawro, A. Blachucik, K. Winiarczyk, K. Litwinowicz, M. Sobański, J. Bladowska, A. Zimny; Wroclaw/PL
Purpose: To examine the association between optic nerve lesion volume (ONLV) on double inversion recovery (DIR) MRI and other imaging biomarkers of disease burden in multiple sclerosis (MS), including white matter hyperintensities (WMHs), T1-weighted hypointensities, and brain volumetrics. The study explores whether ONLV reflects a more severe neurodegenerative profile and may serve as a marker of disease severity.
Methods or Background: In this cross-sectional study, 212 MS patients underwent 3T MRI including 3D T1-weighted, FLAIR, and DIR sequences. Optic nerve lesions were manually segmented on DIR and quantified volumetrically. Patients were categorized by optic nerve involvement: none (n = 59), unilateral (n = 60), or bilateral (n = 93). WMHs were segmented and anatomically classified using an AI-based tool. T1W hypointensities and brain volumetrics were extracted using FreeSurfer and a machine-learning algorithm applied to 3D T1W and FLAIR images. Models were adjusted for intracranial volume.
Results or Findings: ONLV positively correlated with periventricular (r = 0.365, p < 0.001), deep (r = 0.165, p = 0.005), and juxtacortical (r = 0.163, p = 0.007) WMHs. Bilateral involvement was associated with higher WMH burden, increased T1W hypointensities (β = 10.91, p < 0.001), and greater white-matter atrophy (β = –107.02, p = 0.016), particularly along visual pathways. Periventricular WMHs also correlated with global cortical and subcortical gray-matter loss.
Conclusion: ONLV is associated with greater lesion load and neurodegeneration in MS, supporting its potential as a biomarker of disease severity. Longitudinal studies are needed to confirm prognostic value.
Limitations: Cross-sectional design limits assessment of longitudinal progression; future studies should track ONLV evolution and its radiological/clinical impact.

Manual segmentation is prone to variability; AI-based tools could improve accuracy and reproducibility.

Integrating comprehensive clinical data and more uniform cohorts would strengthen future investigations.
Funding for this study: No funding was provided for this study.
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: The study was conducted according to the guidelines of the Declaration of Helsinki and approved by the Institutional Review Board of the Wroclaw Medical University Ethics Committee for conducting research involving humans.

Author Block: H. Larassati¹, S. M. Sceppacuercia², A. H. M. E. Hammam¹, O. Sarwani¹, W. Brownlee¹, C. Auger², A. Rovira Cañellas², T. A. Yousry¹; ¹London/UK, ²Barcelona/ES
Purpose: To evaluate the benefits of spinal cord MRI in multiple sclerosis (MS) by assessing lesion dynamics over time, their relationship with brain lesions, and disability progression.
Methods or Background: This retrospective, multicentre longitudinal study included MS patients from London (UK) and Barcelona (Spain) who underwent brain and spinal cord MRI at baseline and follow-up. Clinical data included relapse history, Expanded Disability Status Scale (EDSS), and disease-modifying therapy (DMT) use. MRI reports were extracted from PACS, with independent reads by two neuroradiologists to assess inter- and intra-observer reliability. Regression models evaluated associations with disability.
Results or Findings: 127 patients (75.6% females, 24.4% males) were included, mean age of 43 years old, median follow-up interval at 19 months (11–29 months), 56.8% were on DMT at follow up. 82.7% of patients had spinal cord lesions at baseline, and 15.1% developed at least one lesion at follow-up. New spinal cord lesions were most frequent at 11–12 months (28.6%) and less common at later follow-ups (8–22%). New spinal cord lesions were more common in the relapsing MS, and was significantly associated with EDSS worsening (p=0.036). In progressive MS, new lesion incidence was lower and showed weaker correlation with disability progression. Spinal-only activity was present in 7.9% patients at baseline and occurred in additional cases at follow-up. Inter- and intra-rater agreement for lesion burden and new lesion detection was moderate to high (κ=0.60–0.83; ICC=0.56–0.94).
Conclusion: New spinal cord lesions predict disability worsening and reveal spinal-only activity missed on brain MRI. New spinal lesion accrual was relatively low and the benefit of spinal MRI may vary by patient phenotype and timing. A selective, phenotype- and timing-based approach to spinal cord imaging may optimise its use in MS monitoring.
Limitations: Small sample size, limited follow-up period
Funding for this study: None
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: University College London Hospitals NHS Foundation Trust

Author Block: B. Atalay, M. Yamer, M. B. Doğan, I. Aydın Cantürk; Istanbul/TR
Purpose: To investigate the association of paramagnetic rim lesion subtypes—complete, incomplete, patch-like, which serve as biomarkers of chronic active lesions in multiple sclerosis—with the Expanded Disability Status Scale (EDSS) and multiple sclerosis subtypes.
Methods or Background: This retrospective study included patients with a confirmed diagnosis of MS who underwent brain MRI. PRLs were evaluated by count, type (complete rim, incomplete rim, or patch-like) and location using phase images from SWI sequences. The presence of active lesions was also noted. Statistical analyses examined associations between PRL characteristics and clinical parameters using non-parametric tests and Spearman correlations.
Results or Findings: Thirty-two patients (75% female; mean age 46.1 ± 15.7 years) were included. PRLs were present in 62.5% of patients, with 28.1% complete and 12.5% incomplete rims, and the remainder were patch-like. PRL count, type, and location did not differ significantly across MS subtypes (p = 0.93, 0.21, and 0.08, respectively). Patients with active lesions exhibited a significantly higher number of PRLs compared to those without (p = 0.02). However, PRL location and type were not associated with the presence of active lesions (p = 0.07 and p = 0.15). No significant correlations were observed between PRL count and either EDSS score or disease duration (p = 0.83 and p = 0.87). Likewise, EDSS score and disease duration did not significantly differ between patients with and without PRLs (p = 0.23 and p = 0.14, respectively).
Conclusion: PRLs are common in patients with MS and are associated with the presence of active lesions, but not with MS subtype, EDSS score, or disease duration. Their role in clinical prognosis should be clarified in larger, prospective studies.
Limitations: The limitations of our study include its retrospective design, the relatively small patient sample size.
Funding for this study: No funding for this study.
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: Our study was approved by an ethics committee.

Author Block: S. Hannoun, G. Fayad, N. El-Ayoubi, S. Khoury; Beirut/LB
Purpose: To investigate whether lesion filling improves the accuracy and interpretability of brain age estimation in multiple sclerosis (MS), and to assess its associations with clinical disability and structural MRI markers.
Methods or Background: We retrospectively analyzed 571 relapsing-remitting MS patients. Brain age was estimated using the BrainAgeR pipeline on both non-lesion-filled and lesion-filled T1-weighted images. Bias correction was applied to adjust for age-related prediction bias, and the Brain Age Gap (BAG) was computed as the difference between corrected predicted age and chronological age. Agreement between approaches was assessed using Bland–Altman analysis, Pearson correlation, and intraclass correlation coefficient (ICC). Associations of BAG with disability measures (EDSS, 9HPT, SDMT, 25FWT) and volumetric MRI metrics (global and subcortical volumes) were evaluated using multivariable regression with Bonferroni correction.
Results or Findings: Lesion-filled and non-lesion-filled brain age estimates showed excellent agreement (r=0.97, ICC=0.962), with a mean difference of 1.23 years. The mean absolute error was slightly lower for lesion-filled predictions (8.12 vs 9.40 years). Both BAG measures were significantly associated with EDSS, 9HPT, and SDMT (p<0.001), but not with 25FWT. Lesion-filled BAG demonstrated stronger associations with gray matter, thalamic, and hippocampal volumes, with higher explained variance compared to non-lesion-filled BAG. These associations remained significant after multiple comparison correction.
Conclusion: Brain age estimation is robust to lesion effects in MS, with lesion filling offering modest improvements in alignment with structural imaging markers but limited impact on clinical correlations. Lesion correction should be considered when precise structural interpretability is required.
Limitations: This cross-sectional, single-center study included only relapsing-remitting patients without healthy controls. Scanner variability and potential circularity between lesion-filled volumetrics and brain age may have influenced associations. Longitudinal, multi-center studies are warranted.
Funding for this study: None
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: This study was approved by the American University of Beirut Biomedical Institutional Review Board (IRB), which explicitly waived the requirement for written informed consent due to the retrospective nature of the study and minimal risk to participants.

Author Block: S. Pasumarthi Venkata¹, C. Arnold¹, P. Gulaka¹, S. Colombo Serra², G. Erb³, G. D'Anna⁴, A. Shankaranarayanan¹, G. Zaharchuk⁵; ¹Menlo Park, CA/US, ²Colleretto Giacosa/IT, ³Konstanz/DE, ⁴Milan/IT, ⁵Stanford, CA/US
Purpose: In the treatment of Multiple Sclerosis (MS), contrast-enhanced (CE) images are crucial in differentiating between new and chronic lesions. Recently, an FDA-cleared contrast boosting (CB) deep learning algorithm was developed that boosted contrast signals present in T1w standard contrast-enhanced (SC) images, improving lesion visualization without increasing dosage. In this work, we evaluate the clinical performance of the CB algorithm on a public dataset consisting of T1w pre and CE images from MS patients.
Methods or Background: From the Open MS Dataset, T1w-pre and SC images from 30 patients (23 Females; 39±10) were used in this study. Contrast boosted (CB) images were generated from T1w-pre and SC using the CB algorithm. Three board-certified radiologists were asked to score the SC and CB images for lesion contrast enhancement, border delineation and internal morphology on a 4-point Likert scale. Readers also scored on any False Lesions (FL) found on CB images and their impact on diagnosis. Contrast-to-noise ratio (CNR), lesion-to-brain ratio (LBR) and contrast-enhancement-percentage (CEP) were computed on SC and CB images.
Results or Findings: The CB images were rated higher (p<0.05) than SC images for enhancement (3.66±0.56 vs 2.25±0.84), delineation (3.62±0.71 vs 2.25±0.84) and morphology (3.41±0.77 vs 2.12±0.89). CB images had higher (p<0.01) CNR, LBR and CEP. The readers found a few enhancing lesions that were almost missed on SC but were clearly visualized on CB. 12% of cases had FLs that could potentially impact diagnosis, but could be ruled out using T2-FLAIR.
Conclusion: We showed that the CB algorithm has superior clinical performance on MS lesions when compared to SC images. The CB algorithm clearly visualized a few lesions that were almost undetectable from the SC images.
Limitations: This study is limited to a small number of cases.
Funding for this study: n/a
Has your study been approved by an ethics committee? Not applicable
Ethics committee - additional information:

Author Block: O. Alomair¹, S. A. Alghamdi¹, A. abujamea², M. S. Alshuhri³, S. Aljarallah¹, N. Alkhawajah¹, H. Al-Mubarak⁴, Y. Alashban¹, N. Kurniawan⁵; ¹Riyadh/SA, ²riyadh/SA, ³Al Kharj/SA, ⁴Glasgow G61 1QH/UK, ⁵Brisbane QLD 4072/AU
Purpose: In this abstract, we achieved three aims previously published in three papers. First, evaluate intravoxel incoherent motion (IVIM) diffusion and perfusion MRI metrics for various types of MS lesions, including enhanced, non-enhanced, and black hole lesions. Second, investigate the correlation and predictive values of the IVIM diffusion and perfusion MRI metrics with disability status. Third, utilise radiomics features to evaluate the predictive value of IVIM diffusion parameters in relation to disability severity.
Methods or Background: This cross-sectional study retrospectively analysed quantitative IVIM parameters and MRI data from 197 MS patients. Multiple linear regression was applied to identify independent predictors of EDSS score. Machine learning (ML) techniques, such as XGB, Random Forest, and ANN, were employed to explore the relationships between radiomic IVIM and clinical variables.
Results or Findings: In this abstract, we presented the results previously published in three papers. First, ADC, D, and D* values for MS black hole lesions were significantly higher (p < 0.0001) than those for other MS lesions. Second, in the multivariate regression analysis, only the number of MS lesions and relapses emerged as independent predictors of EDSS score (p-value < 0.001). Third, for disability prediction, IVIM-D and D* radiomics strongly correlated with EDSS: Random Forest achieved 89% accuracy (AUC = 0.90), while CNN achieved 90% accuracy (AUC = 0.95).
Conclusion: These three published studies demonstrate the utility of IVIM parameters in detecting microstructural alterations associated with MS impairment. Machine learning analyses of IVIM metrics provided independent predictors of functional impairment and disability in MS. It validated our results.
Limitations: This study has several limitations, which include a single time point study, and it was limited to analysis of MS lesions without considering normal-appearing white or grey matter.
Funding for this study: This research was funded by the King Salman Center for Disability Research through Research Group no. KSRG-2024-197.
The presented work based on three published paper; Investigating the Role of Intravoxel Incoherent Motion Diffusion-Weighted Imaging in Evaluating Multiple Sclerosis Lesions, The Utility of Intravoxel Incoherent Motion Metrics in Assessing Disability in Relapsing–Remitting Multiple Sclerosis and IVIM-DWI-Based Radiomics for Lesion Phenotyping and Clinical Status Prediction in Relapsing–Remitting Multiple Sclerosis.
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: This study was conducted in accordance with the Declaration of Helsinki and approved by the Local Ethics Committee from King Saud University, Medical City (No. E-23-7517; approval date—22 January 2023; date of renewal of ethical certificate—30 June 2025).

Author Block: L. Wang, C. Arnold, Z. Zhou, L. Xiang, A. Shankaranarayanan, S. Bash, L. N. Tanenbaum, S. Pasumarthi Venkata; Menlo Park, CA/US
Purpose: Double inversion recovery(DIR) MRI provides superior gray–white matter differentiation and lesion visibility in multiple sclerosis(MS), but is rarely acquired in routine practice due to long scan times. We repurposed a synthesis model originally developed for STIR and applied it to multiple brain MS datasets to evaluate its cross-domain generalization. Despite being trained in a different anatomical setting, the model demonstrates strong performance in generating DIR-like (Syn-DIR) images, with consistent improvements across multisite MS datasets.
Methods or Background: The model was directly applied to T1-weighted and T2-FLAIR brain scans to generate DIR-like images without additional training. Two external datasets were analyzed: open_ms_data(30 subjects) and MSLesSeg(75 subjects). Evaluation included structural fidelity (SSIM and Dice scores on brain volume analysis), tissue contrast (gray-to-white-matter ratio and lesion-to-white-matter ratio), multisite consistency analysis, and longitudinal stability analysis.
Results or Findings: FLAIR and SynDIR showed high structural fidelity across sites (mean SSIM: 0.8719 ± 0.0217), with Dice scores >0.85 for all major brain structures.
SynDIR images exhibited significantly improved tissue and lesion contrast compared with FLAIR. Specifically,gray-to-white-matter ratio increased from 1.2688 ± 0.0617 to 1.6134 ± 0.1946, and lesion-to-white-matter ratio increased from 1.4035 ± 0.1309 to 1.9335 ± 0.3833 (p < 0.001).
The t-test on the ratio of gray-to-white-matter ratio to lesion-to-white-matter between the two cohorts showed no significant difference, indicating consistent performance across sites. In addition, the Wilcoxon signed-rank test showed no significant difference between time points (p = 0.1478).
Conclusion: A spine-trained synthesis model generalized effectively to brain MS cohorts, generating DIR-like images that enhanced lesion visibility and showed consistent performance across sites and timepoints. This repurposed approach enables advanced contrasts for MS studies without acquiring the DIR series.
Limitations: In future work, these methods should undergo qualitative review by radiologists.
Funding for this study: n/a
Has your study been approved by an ethics committee? Not applicable
Ethics committee - additional information:

Author Block: E. Masiello¹, A. Diena¹, G. M. Agazzi², A. Falini², N. E. Anzalone²; ¹Milan/IT, ²Milano/IT
Purpose: In multiple sclerosis (MS) MRI is essential for diagnosis and follow-up. Detecting new or enlarging lesions can be challenging and reporting is time-consuming. This study aimed to evaluate the role of an AI platform in lesion detection compared with neuroradiologists of different experience levels, while also assessing its impact on reading time.
Methods or Background: This retrospective observational study included 59 adult patients with relapsing–remitting MS who underwent two brain MRI examinations with standardized 3D-FLAIR sequences at baseline and follow-up. Two neuroradiologists with >10 years and 3 years of experience independently reviewed anonymized scans for new or enlarging lesions, blinded to AI output and each other. The AI software (TensorMedical, Spain) automatically quantified new and slowly expanding lesions by co-registering longitudinal scans. Consensus between the two radiologists was considered the ground truth. Diagnostic accuracy, Cohen’s kappa agreement, and reading times with and without AI support were analyzed.
Results or Findings: The cohort included 40 females and 19 males (mean age=43.2 years). The average time interval between baseline and follow-up MRI was 14.2 months (range 2.3–51.1). The mean difference in lesion count compared to ground truth was 0.322±0.797 for the senior reader (N.A), 0.373±1.410 for the junior reader (A.D), and –0.763±4.440 for AI. Overall differences across groups were statistically significant (ANOVA, p=0.0409), but post-hoc comparisons didn't reach significance. Agreement with ground truth was almost perfect for the senior reader (κ=0.898), substantial for the junior reader (κ=0.695), and substantial for AI (κ=0.797). AI assistance reduced median reading time from 5.5 to 2.0 minutes
Conclusion: AI performance was comparable to that of an experienced neuroradiologist and superior to a junior reader. Integration of AI reduced reading time and enhance consistency, particularly for less experienced neuroradiologists.
Limitations: Retrospective, single-center, limited sample size.
Funding for this study: No funding.
Has your study been approved by an ethics committee? Not applicable
Ethics committee - additional information: