Background/Purpose: There is growing interest in the management of early axial spondyloarthritis and intervention with disease-modifying
agents while disease is still confined to the sacroiliac joints (SIJ). Assessment
of structural lesions of the SIJ on MRI may be a helpful tool to monitor
disease progression but requires evidence that change in such lesions can be
reliably detected and the type of calibration necessary to achieve this. We
aimed to assess reliability of detection of structural lesions in the SIJ on
MRI and the impact of calibration using either a standardized web-based training
module or a set of DICOM-based reference MRI scans.

Methods: In this international multicenter study, 5 readers without exposure to the application of scoring
methods specifically for structural lesions in the SIJ reviewed
either a web-based training module (n=3) that included standardized definitions
for and examples of erosion, fat metaplasia, backfill, and ankylosis,
or a set of DICOM-based reference MRI scans (n=2). Both calibration methods
included instructions for the scoring of lesions according to the Spondyloarthritis Research Consortium of Canada SIJ
Structural Score (SSS) on 5 consecutive semi-coronal slices through the SIJ using
a binary method based on presence/absence of the lesion in SIJ quadrants
(erosion, fat metaplasia (range 0-40)) or SIJ halves (backfill, ankylosis range (0-20)).  Baseline and 2-year T1-weighted (T1W)
scans from 30 patients with axial SpA blinded to patient
characteristics, short-tau inversion recovery (STIR) scan, and time point, were
scored and data entered directly online using a web-based schematic of the SIJ.
Interobserver reliability for status and change
scores was calculated by intra-class correlation coefficient (ICC) and
comparisons made with pre-specified expert readers (n=3).

Results: Mean (SD) reduction in SSS erosion score was significantly greater
for expert readers (-1.62 (4.05)) and for DICOM-trained readers (-1.40 (3.42))
than for web-based module trained readers (-0.47 (1.56)). Mean (SD) increase in
SSS backfill score was significantly greater for expert readers (0.82 (3.87))
than for web-based module trained readers (-0.07 (1.67)). ICC differed between
groups mainly for 2-year change in erosion and backfill. Although reliability
was superior for DICOM-trained readers, substantial reliability with expert
readers was attainable following training with the web-based module (Table).

Conclusion: Structural lesions in the SIJ on MRI are often heterogeneous in
appearance and their evolution may not be reliably detected without the more
rigorous approach to calibration using DICOM scans.  

Table. Interobserver ICC

*ICC for each reader based on comparison
with mean score of 3 expert readers