Initially, 32 patients with ischemic stroke aged 60 years or younger were screened for participation. Of these, 13 were excluded because they had only the index ischemic stroke lesion without additional white matter lesions on MRI, or had lesions that did not meet the inclusion criteria, resulting in a final sample of 19 stroke patients. The stroke cohort consisted of patients examined in a non-acute phase after the index ischemic stroke (median time from stroke to MRI 507 days; IQR 339–849 days). In addition, 30 patients with clinically confirmed multiple sclerosis (MS) were included, yielding a total study population of 49 participants.

All participants underwent a single MRI examination at Uppsala University Hospital between 2021 and 2024, utilizing a standardized imaging protocol. Inclusion criteria for all patients included, being ≥ 18 years of age and availability of a complete MRI imaging including fluid-attenuated inversion recovery (FLAIR) and synthetic MRI sequences. For the MS cohort, demographic data, expanded disability status scale scores (EDSS), and data on disease duration, time since last relapse, MS subtypes, and distribution of disease-modifying therapies (DMTs) were collected. EDSS was presented as median (interquartile range, IQR) (Table 1).

As for the patients with ischemic stroke, data on demographic parameters, vascular risk factors, cardiac comorbidities, and stroke subtypes were retrieved from medical records (Table 2).

All MRI scans were performed on a 3T scanner (Achieva dStream, Philips Medical Systems, Best, The Netherlands) equipped with a 32-channel head coil. The imaging protocol included a three-dimensional FLAIR sequence and a multi-dynamic multi-echo (MDME) sequence optimized for synthetic MRI acquisition. In the stroke cohort, diffusion-weighted imaging (DWI) was additionally available and was reviewed for lesion characterization and exclusion of acute ischemic lesions.

Specific sequence parameters were as follows: FLAIR (TR = [4800] ms, TE = [304] ms, TI = [1650] ms, slice thickness = [1.1] mm) and MDME (TR = [4688] ms, TE = [12.5] ms, slice thickness = [4] mm). The MDME sequence enabled the generation of quantitative maps of R1, R2 relaxation rates, proton density (PD), and myelin content using the SyMRI software versions (7.1, 2017; 7.3, 2018; 8.0, 2019;11.0.7, 2019; 12.1.11, 2024) Synthetic MR AB, Linköping, Sweden).

In addition to quantitative parameter maps, SyMRI reconstructs synthetic contrast-weighted images (including T1-, T2-, and FLAIR-weighted images) derived directly from the same quantitative dataset, ensuring inherent spatial correspondence between the anatomical images and the quantitative maps.

Criteria for typical MS lesions included periventricular or corpus callosum location, ovoid shape, perpendicular orientation of the long axis towards the adjacent lateral ventricle, and size > 5 mm.

For each patient, representative non-specific white matter lesions were initially identified on the conventional FLAIR sequence, which served as the anatomical reference for lesion detection. Non-specific lesions were defined as FLAIR-hyperintense lesions lacking typical morphological features of MS lesions and meeting the predefined inclusion criteria: lesion size > 5 mm in maximum diameter, located at least 10 mm from the ventricular system and 5 mm from the cortical surface (outer edge of FLAIR hyperintensity), absence of contrast enhancement, and lesions not appearing hypointense on T1-weighted sequences.

In the stroke cohort, diffusion-weighted imaging (DWI) was reviewed to confirm the absence of acute ischemic lesions. Lesions demonstrating diffusion restriction were therefore not included in the analysis. In addition, lesions with imaging characteristics typical of infarction, such as those corresponding to a defined arterial vascular territory or demonstrating cerebrospinal fluid–like signal compatible with chronic lacunar infarction, were excluded.

After lesion identification on conventional FLAIR images, the corresponding lesions were visually matched to the synthetic FLAIR images generated within the SyMRI software environment. Regions of interest (ROIs) were subsequently delineated on the synthetic FLAIR images. Because the synthetic images are reconstructed directly from the same quantitative dataset, they are inherently spatially aligned with the quantitative maps, and no additional anatomical co-registration step was required. Once the ROIs were placed, the SyMRI software automatically extracted the corresponding quantitative parameters (R1, R2, proton density, and myelin content) from the underlying maps.

During ROI placement, care was taken to maintain a sufficient margin between the ROI borders and the lesion edges as well as adjacent anatomical boundaries to minimize partial volume effects. Only lesions providing adequate surrounding white matter space for reliable ROI placement were included in the analysis.

In addition to lesion analysis, one ROI was placed in normal-appearing white matter (NAWM) for each of the 30 MS and 19 stroke patients. NAWM was visually identified on conventional and SyMRI-generated synthetic FLAIR images, as white matter regions without visible abnormalities. ROIs were placed in visually normal regions at a sufficient distance from focal lesions to minimize partial volume effects.

Quantitative evaluation of the selected lesions was conducted using the SyMRI post-processing software. For each ROI, R1, R2, proton density, and myelin content (MyC) were extracted. In addition to lesion analysis, a ROI was also placed within normal-appearing white matter (NAWM) to obtain reference measurements for comparison. Lesion selection was initially performed by a first rater who was not blinded to the patient group (MS or stroke). A second rater, an experienced neuroradiologist, reviewed all lesions under blinded conditions; however, full blinding could not always be ensured when overt radiological features suggested the underlying diagnosis.

For all analyses, the mean value for the lesions was calculated at the patient level. Comparisons between typical MS lesions (MSL) and non-specific lesions in MS patients (nsWML-MS) were performed using paired Wilcoxon signed-rank tests, while differences between nsWML-MS and non-specific lesions in stroke patients (nsWML-S) were assessed using Mann–Whitney U tests. Logistic regression models, both unadjusted and adjusted for age, were applied to MyC, PD, R1, and R2 values to predict MS status, with results reported as odds ratios (OR) and 95% confidence intervals (CI). Receiver operating characteristic (ROC) curves were generated to evaluate discriminatory ability. Comparisons of normal-appearing white matter (NAWM) between MS and stroke groups were also conducted using Mann–Whitney U tests and age-adjusted logistic regression.