Quantitative techniques for lesion load measurement in multiple sclerosis: an assessment of the global threshold technique after non uniformity and histogram matching corrections.

Abstract

The quantitative assessment of abnormalities on brain MRI (magnetic resonance imaging) in multiple sclerosis (MS) provides a widely used endpoint in monitoring treatment efficacy. Semi-automated intensity thresholding (global thresholding) is an established method of lesion segmentation, but it is limited by the heavy dependence of derived volumes on the chosen intensity threshold. Inconsistency in threshold choice substantially contributes to its suboptimal reproducibility. A potential improvement to global thresholding is to utilize the histogram matching algorithm to correct for variations in scanner sensitivity. This study evaluated whether or not a single intensity threshold could then be applied across multiple histogram matched images. Eight MS patients were scanned at baseline and after nine months. After non-uniformity correction, the histogram matching correction was applied, reducing the mean absolute percentage variation in normal appearing white matter signal intensity from 15.2 to 3.8% for baseline scans and from 11.9 to 2.1% across serial studies. Lesion volumes were measured with the global threshold technique using a single threshold selected from one scan. A local thresholding technique (contouring) was also performed as a gold standard measure. Agreement between techniques for baseline lesion volumes was only moderate with an intra-class correlation coefficient (ICC) of 0.46 and there was little agreement for change in lesion volume (ICC = 0.17), reflecting inconsistencies in the volume of non-lesion regions included by the threshold. Agreement was improved by deleting non-lesion regions, but this was time consuming and less reproducible than contouring method. At present, the contour technique remains a more appropriate method for lesion load quantification.