Image registration and subtraction to detect active T₂ lesions in MS: an interobserver study

Abstract.

Serial MRI studies are used to analyse change in multiple sclerosis (MS) lesion volume in clinical trials. As such an evaluation is very time consuming and subject to quantification errors, one might assess only the change in number or size of lesions using subtracted images. The advantage of subtracted images is that both new and/or enlarging and resolving and/or shrinking lesions can be evaluated, resulting in a more precise volume change than a net volume change. We studied the interobserver agreement in the detection of active MS lesions using paired dual-echo T₂-weighted spin-echo studies (3-mm slices) of 30 MS patients with a range of MS disease activity on MRI from treatment trials. Using an automatic matching algorithm based on mutual information, the follow-up scan was registered to baseline, after which subtracted images were obtained. After a training session with formulation of guidelines, six observers identified new, enlarging, resolving and shrinking lesions on subtracted images. Weighted kappa (κ) values were calculated to assess interobserver agreement. Good agreement was found for new lesions (κ 0.69 ± 0.08), while moderate agreement was found for enlarging lesions (κ 0.52 ± 0.06). When new and enlarging lesions were combined, good agreement was found for “positive” activity (κ 0.71 ±0.06). The interobserver agreement was poor for resolving lesions (κ 0.31 ± 0.07), and moderate for shrinking lesions (κ 0.53 ± 0.08). In conclusion, the use of subtracted images in the visual detection of new T₂ lesions resulted in a good level of interobserver agreement for “positive” disease activity. Subtraction of registered images is a reliable, time efficient method to assess disease progression in MS.