Abstract
Scoring erosions on magnetic resonance imaging (MRI) is one method of estimating damage in patients with rheumatoid arthritis (RA), but it has limitations. The aim of this pilot study was to assess the feasibility and inter-reader reliability of computer assisted erosion volume estimation in patients with RA. Intra-reader and inter-occasion reliability was also assessed, and different slice thicknesses were compared in terms of erosion volume estimation. A 3 mm slice thickness 3D gradient-echo sequence followed by a 1 mm sequence was performed at baseline and repeated within 24 h with metacarpophalangeal (MCP) joints 2 to 5 of the dominant hand included in the field of view. Three readers were instructed to grade MCP 2 and 3 using the OMERACT grading system and then to measure the erosion volume of the same joints using OSIRIS software. The inter-reader reliability of the grading method and the volume method was calculated, as well as the inter-occasion reliability, by comparing results from each reader from baseline to the followup scan. One reader performed repeat volume measurements on 5 patients to assess the intra-reader reliability. Five patients were included in the study. Expressed in terms of intraclass correlation coefficients (ICC), the inter-reader and inter-occasion reliability of the volume method were comparable to the existing OMERACT scoring system, but large systematic differences in volume estimations were found between readers. The intra-reader reliability was excellent. Good correlation was demonstrated between the total erosion scores and the total erosion volumes. For both erosion volumes and erosion scores, 1 mm and 3 mm acquisitions produced variable results between readers, with no clear pattern of underestimation or overestimation for either slice thickness. The volume estimation method was more time consuming, taking roughly 5 times as long as the scoring method. Computerized MRI erosion volume measurements are feasible, with high intra-observer and inter-occasion reliabilities. Despite high ICC, the inter-observer reliability is not sufficient for multicenter use without prior reader training and calibration. The optimal slice thickness was not determined.
In this issue
