Evaluation of the use of segmentation-based and deconvolution-based three-dimensional joint space width analysis of the hand and wrist

Abstract Three-dimensional morphometric analyses of joint space from computed tomography (CT) scans can provide a quantitative and unbiased evaluation of radiographic damage from arthritis. This potentially offers greater sensitivity for disease assessment and monitoring than traditional ordinal scoring systems. Previously established methods for three-dimensional joint space analysis from CT scans can be categorized into segmentation-based and grayscale deconvolution-based methods. However, segmentation-based methods have not been employed for joints other than metacarpophalangeal joints, nor have they been compared to deconvolution-based methods. This study compared the joint space width (JSW) distribution from various hand and wrist joints in photon-counting CT images, using both measurement approaches. From these distributions, the mean (JSW.Mean), standard deviation (JSW.SD), maximum (JSW.Max), minimum (JSW.Min), 95th percentile (JSW.95%), and 5th percentile (JSW.5%) values were calculated. For each morphometry parameter, agreement between the two methods was assessed by calculating the coefficient of determination (R2), and the mean and standard deviation of the individual biases. General estimating equations were used to assess the relationship between morphometry and an equivalent/modified Sharp/van der Heijde (SvdH) ordinal radiographic score. In general, agreement between the two methods was low: R2 ranged between 0.42 and 0.87; biases up to 0.18 mm $\pm$ 0.33 mm). For both methods, an increased degree of joint space narrowing as assessed with SvdH was associated with reduced JSW.Min and increased JSW.SD; for the deconvolution method, it was also associated with decreased JSW.Mean and increased JSW.Max. Interactions between the methods showed a weaker association to SvdH score for JSW.Min and a stronger association for JSW.SD when measured with the deconvolution method, compared to the segmentation method. We conclude that both methods lead to fundamentally different measurement outcomes, and that the deconvolution method might be more sensitive to radiographic damage.