Lesion assessment in multiple sclerosis: a comparison between synthetic and conventional fluid-attenuated inversion...

Magnetic resonance imaging (MRI)-based lesion quantification is essential for the diagnosis of and prognosis in multiple sclerosis (MS). This study compares an established software's performance for automated volumetric and numerical segmentation of MS brain lesions using synthetic T2-weighted fluid-attenuated inversion recovery (FLAIR) MRI, based on a multi-dynamic, multi-echo sequence (MDME), vs. conventional FLAIR imaging.

To ensure comparability, 3D FLAIR images were resampled to 4 mm axial slices to match the synthetic images' slice thickness. Lesion segmentation was performed using the Lesion Prediction Algorithm within the Lesion Segmentation Toolbox. For the assessment of spatial differences between lesion segmentations from both sequences, all lesion masks were registered to a brain template in the standard space. Spatial agreement between the two sequences was evaluated by calculating Sørensen-Dice coefficients (SDC) of the segmented and registered lesion masks. Additionally, average lesion masks for both synthetic and conventional FLAIR were created and displayed as overlays on a brain template to visualize segmentation differences.

Both total lesion volume (TLV) and total lesion number (TLN) were significantly higher for synthetic MRI (11.0 ± 12.8 mL, 19.5 ± 12.1 lesions) than for conventional images (6.1 ± 8.5 mL, 17.9 ± 12.5 lesions). Bland-Altman plot analysis showed minimal TLV differences between synthetic and conventional FLAIR in patients with low overall lesion loads. The intraclass coefficient (ICC) indicated excellent agreement between both measurements, with values of 0.88 for TLV and 0.89 for TLN. The mean SDC was 0.47 ± 0.15.

Despite some limitations, synthetic FLAIR imaging holds promise as an alternative to conventional FLAIR for assessing MS lesions, especially in patients with low lesion load. However, further refinement is needed to reduce unwanted artifacts that may affect image quality.