Automated measurement of joint space width in early rheumatoid arthritis hand radiographs

Abstract

Rheumatoid arthritis (RA) is a chronic inflammatory disease, affecting predominantly small joints of hands and feet. The current gold standard for assessment of radiographic progression in RA is the Sharp/van der Heijde scoring method (SvdH), scoring both bone erosions and joint space narrowing (JSN). The SvdH scoring method is observer-dependent, and its scores are on an ordinal scale with skewing of the distribution of scores towards 0. These aspects limit the sensitivity of SvdH scoring, which is increasingly an issue in the light of limited radiographic progression in early RA nowadays. The aim of the research presented in this thesis was to develop an automated system for radiographic hand JSW measurement in early RA, for an objective and potentially more sensitive JSN assessment compared to the conventional scoring methods. Firstly, two methodological differences between conventional and automated methods in JSN assessment were investigated, viz. the difference in the joints to be assessed and the difference in the reading strategies. It was shown that the influence of these differences to JSN (progression) assessment was limited. Subsequently, automated methods for quantification of radiographic joint space width (JSW) of fingers and wrist were presented, respectively, including joint location detection, margin delineation and JSW measurement. Based on the proposed methods, a software package ‘Joint Space Width Quantification’ (JSQ) was developed and evaluated in a sub-group of a treatment strategy trial in early RA, in which no statistical differences in JSN progression were observed using SvdH scoring. Quantified using JSQ, the rate of yearly JSW change of the hands was higher in the group initiating placebo in addition to conventional treatment than in the group initiating prednisone in addition to conventional treatment. Moreover, the effect size was larger when using JSQ than SvdH scoring. This proof of concept study shows that the automated software package JSQ is sensitive to JSW change of the hand in an early RA clinical trial, in which only limited radiographic progression was present. As is known, assessment of radiographic progression is prone to variation due to differences in positioning during radiograph acquisition. Possible benefits of using a hand position-fixing mould during radiograph acquisition were thus investigated, both for the conventional SvdH scoring method and for the automated hand JSW measurement by JSQ. It was shown that using the hand mould resulted in similar reliability and better repeatability of joint damage assessment for SvdH, but not for automated JSW measurement by JSQ. However, with the use of the hand mould, a slight improvement was observed for the repeatability of the automated JSW measurement of the proximal interphalangeal joints (PIPs). To summarize, an automated system for hand JSW measurement has been developed and its value on the quantification of hand JSN in early RA has been preliminarily validated. It is recommended to incorporate the system into quantitative assessment of radiographic progression of RA in clinical trials.