Segmentation of Subtraction Images for the Measurement of Lesion Change in Multiple Sclerosis

Yin-Yang artifact: scan-rescan MR images from the same day. Left column, Baseline image (scan). Middle column, Coregistered second time-point image (rescan). Right column, Subtraction image (rescan-scan). Yin-Yang artifacts are shown (white arrows) in the subtraction image (right column). An artifact from vessel misregistration is shown (white arrowhead) in the subtraction image. (In all of the images, the skull has been stripped by masking with ICC.)

Positive, negative, and net changes from pairs of subtraction images: the discriminatory power of SSEG. Change measurements obtained from pairs of dual-echo images using CSEG and SSEG methods are depicted. A, Patient-by-patient representation of lesion volume changes measured with SSEG and CSEG in pairs of dual-echo MR images from 21 patients with MS. The CSEG method can only measure the net change in lesion volume (black bar in each patient). In contrast, the SSEG method provides information about the volume change in new (red bars), enlarging (yellow bars), and resolving lesions (blue bars), as well as yielding a net lesion volume change (sum of new and enlarging lesion volumes minus resolving lesion volume; green bars). B, The average lesion volume changes measured with the CSEG and SSEG methods are shown. Greater average net lesion volume change is measured with the SSEG method, although no statistically significant difference is found in the net lesion volume change measured with both methods (P = 0.14, Wilcoxon test).

Reproducibility of SSEG and CSEG methods: Bland-Altman analysis in 10 patients (scan-rescan group). The solid line and the 2 dotted lines represent the mean ± 1.96 SD (95% confidence interval) of the difference of rescan lesion volume and baseline lesion volume, respectively. For both measurements, we use the same baseline lesion volume measured by CSEG. A, For the CSEG measurement, we used the baseline and rescan lesion volumes. B, For the SSEG measurement, the rescan lesion volume is the sum of baseline lesion volume (measured by CSEG) and the net change in lesion volume measured by the SSEG method. We see that the SSEG method has smaller confidence intervals, which indicate greater reproducibility.

A juxtacortical lesion in a 44-year-old female patient with MS scanned at an interval of 3 years. A new juxtacortical lesion (arrow), difficult to appreciate on the native images and missed by CSEG, is clearly visible on the subtraction image (C) and SSEG (F). Baseline image (A) and its CSEG (D); Coregistered second time-point image (B) and its CSEG (E). The CSEG images were coregistered in this example to allow direct comparison. The CSEG method (D and E) segments CSF (blue), GM (orange), lesion (yellow), and WM (green); and SSEG method (F) only segments new lesion (pink). Subtle artifacts are seen on the boundary of the brain surface due to slight misregistration. (In all of the images, the skull has been stripped by masking with ICC.)