Quantification of MRI lesion load in multiple sclerosis: a comparison of three computer-assisted techniques

Magn Reson Imaging. 1996;14(5):495-505. doi: 10.1016/0730-725x(96)00018-5.


Several computer-assisted techniques for measuring multiple sclerosis lesion load on MR images have been developed to provide a quantitative and sensitive means for monitoring disease activity, particularly in the context of treatment trials. We have evaluated three techniques: manual outlining (similar to that of the North American interferon beta-1b trial), semiautomated lesion contouring (local lesion based threshold), and intensity-based thresholding for the whole brain. Contiguous, 5 mm-thick, axial, T2-weighted images of the brain were obtained on a 1.5T MR imager in eight patients with clinically definite multiple sclerosis. Analyses of the scans were performed twice, independently by three operators, using the three different techniques. The coefficient of variation of the measurement techniques was: (a) intrarater precision, 9.0 +/- 5.2 (mean +/- SD) (range 0.4-18.5) for the manual outlining, 2.5 +/- 2.1 (0.1-7.7) for the contour technique, and 7.5 +/- 6.9 (0.2-22.0) for the global threshold technique; (b) interrater precision, 11.0 +/- 5.8 (4.9-21.7) for the manual outlining, 4.5 +/- 1.6 (1.8-6.6) for the contour technique, and 11.4 +/- 4.9 (2.8-19.2) for the global threshold technique (0.0 = perfect precision). The absolute lesion loads measured were very similar using the manual outlining and the contour techniques but were significantly smaller using the global threshold technique. We conclude that the contour technique is a promising tool for use in treatment trials. Further studies are needed to assess sensitivity to changes in lesion load over time.

Publication types

  • Comparative Study
  • Research Support, Non-U.S. Gov't

MeSH terms

  • Brain / pathology*
  • Humans
  • Image Processing, Computer-Assisted*
  • Magnetic Resonance Imaging / methods*
  • Multiple Sclerosis / pathology*
  • Observer Variation
  • Reproducibility of Results