Background/Purpose: Instruments designed to determine the clinical relevance of bone erosions captured by magnetic resonance imaging (MRI) in rheumatoid arthritis (RA) patients have been set-back by inconsistency in both intra and inter-rater reliability. The developed Early Erosions in Rheumatoid Arthritis (EERA) software holds promise for increasing reliability across readers by quantifying RA patient erosions in a semi-automated fashion using an amalgamation of conventional Region Growing and Level-Set Segmentation algorithms. The principle aim of this study was to determine intra and inter-rater reliability when applying EERA software to the quantification of erosions in the metacarpal phalanges (MCPs) of RA patients.

Methods: Two readers, R1 and R2, trained to use EERA software, but otherwise inexperienced with conventional quantification techniques, evaluated erosions captured by MRI in the second through fifth MCPs of 50 patients diagnosed with RA under the American College of Rheumatology 1987 revised definition. A 1T magnet, 100mm diameter cylindrical transmit and receive coil, and a 3D spoiled gradient echo sequence were used in acquiring the images. Images were evaluated by each reader twice with a 72 hour wait period between runs. Intra and inter-rater reliabilities for the total volume measures between the two readers and two runs were assessed via intra-class correlations, ICC(2,1), with 95% confidence intervals. For each run, volume differences between readers were graphed against R2’s volume measures in Bland-Altman difference plots so as to visually capture the degree to which scores varied.

Results: Of the 50 participants recruited for the study, 16 were male and 34 were female. Study subjects had a mean age of 57 (SD=11.5)yr, a mean weight of 78 (SD=15.6)kg, and a mean height of 169 (SD=13.9)cm. Readers identified 64 ± 1 erosions in the patients: 15 of these occurred in second MCP, 33 ± 1 in the third MCP, 12 ± 1 in the fourth MCP, and 4 in the fifth MCP. Mean erosion size, as determined by R1 during the first and second runs, were 87.1 (SD=118.9)mm³ and 88.1 (SD=121.2)mm³ respectively. R2’s measures had a mean erosion volume of 90.7 (SD=130.1)mm³ for the first run and 103.2 (SD=151.0)mm³ for the second run. For both runs, agreement between readers was better for smaller sized erosions decreasing appreciably beyond 100mm³ (See Figure 1. Run 2 results are not shown but are available upon request). The intra-rater reliability had an ICC value of 0.956 with the 95% confidence interval ranging from 0.935 to 0.970. Between R1 and R2, the inter-rater reliability had an ICC value of 0.921 with a 95% confidence interval from 0.886 to 0.946.

Conclusion: Results obtained suggest that EERA software can be applied to acquire MCP erosion volume measures in a reliable manner.

Figure 1. Run 1 Reader Measured Erosion Volume Differences Plotted against R2’s Measures