Background and purpose: Quantitative neuroimaging is an important part of multiple sclerosis research and clinical trials, and measures of lesion volume (LV) and brain atrophy are key clinical trial endpoints. However, translation of these endpoints to heterogeneous historical datasets and nonstandardized clinical routine imaging has been difficult. The NeuroSTREAM technique was recently introduced as a robust and broadly applicable surrogate for brain atrophy measurement, but no such surrogate currently exists for conventional T2-LV. Therefore, we sought to develop a fully automated proxy for T2-LV with similar analytic value but increased robustness to common issues arising in clinical routine imaging.
Methods: We created an algorithm to identify salient central lesion volume (SCLV), comprised of the subset of lesion voxels within a specific distance to the lateral ventricles (centrality) and with intensity at least a quantitatively-derived amount brighter than normal appearing tissue (salience). We evaluated this method on four datasets (clinical, inter-scanner, scan-rescan, and real-world multi-center), including 1.5T, 3T, Philips, Siemens, and GE scanners with heterogeneous protocols, to assess agreement with conventional T2-LV, comparative relationship with disability, reliability across scanners and between scans, and applicability to real-world scans.
Results: SCLV correlated strongly with conventional T2-LV in both research-quality (r = .90, P < .001) and real-world (r = 0.87, P < 0.001) datasets. It also showed similar correlations with Expanded Disability Status Scale, as conventional T2-LV (r = 0.48 for T2-LV vs. r = 0.45 for SCLV). Inter-scanner reproducibility (ICC) was 0.86, p < 0.001 for SCLV compared to 0.84, p < 0.001 for conventional T2-LV, whereas scan-rescan ICC was 0.999 for SCLV versus 0.997 for T2-LV.
Conclusions: SCLV is a robust, fully-automated proxy for T2-LV in situations where conventional T2-LV is not easily or reliably calculated.
Keywords: MRI; automated lesion detection; multiple sclerosis; proxy.
© 2019 by the American Society of Neuroimaging.