A new method to measure trabecular bone texture on hand radiographs: data from the osteoarthritis initiative

Abstract

Purpose: Grading of hand radiographs for joint space narrowing and osteophytes is the traditional method for assessing hand OA. However this assessment can be difficult and inaccurate since the changes of cartilage volume in finger joints are small the grading requires experienced readers to be reproducible and there is no sensitivity to early OA. Thus a new hand OA assessment method is required. A solution could be in the applications of fractal analyses of finger bone texture regions selected on hand radiographs. The reasons are that the bone texture changes in early stages of OA it exhibits fractal nature and it is related to 3D bone structure. However currently there is no method that could quantify accurately small size bone regions on hand radiographs. We have developed a new method called augmented variance orientation transform (AVOT) method to measure the roughness and directionality of small texture regions at individual scales. Methods: The AVOT method calculates the fractal signatures (FSs) in different directions. FS is a set of fractal dimensions (FDs) calculated at individual image sizes (i.e. scales) while FD is a measure of texture roughness. High value of FD means a rougher texture. The method allows for the analysis of hand bone texture regions that are small in size and selected on arbitrarily oriented fingers.Initially we evaluated whether our method can accurately differentiate between computer generated isotropic and anisotropic fractal texture images of sizes ranging from 20 × 20 to 64 × 64 pixels. This is because roughness and anisotropy of bone texture changes with OA. These sizes correspond to those found on hand radiographs. Isotropic textures had FDs varying from 2.1 to 2.9 in steps of 0.1 (400 images per FD) while anisotropic textures had dominating directions of 120° and 30° (400 images per direction).For those images three parameters at scales ranging from 2 to 7 pixels (depending on image size) were calculated i.e. FS along the direction of the texture roughest part (FSSta) aspect ratio (StrS) and direction signatures (StdS) respectively. The aspect ratio and direction measure the texture anisotropy. We also evaluated our method for its sensitivity in differentiating the bone texture between subjects with and without radiographic hand OA. The subjects were taken from the Osteoarthritis Initiative (OAI) public use data set (OAI Datasets 0.2.2 and 0.C.1). Images from centre A were used since they exhibit smallest size amongst the five centres (150 DPI). This strategy allows for testing “the worst scenario” i.e. 20 × 20 pixel regions. We used 20 pairs of subjects (n = 20 14 women) with (OA cases) and without (controls) OA in the 5th distal interphalangeal (DIP5) joint. OA was defined as: (i) joint space narrowing (JSN) grade 2 or worse (ii) osteophyte grade 2 or worse or (iii) JSN grade 1 with an osteophyte grade 1. These criteria approximate Kellgren and Lawrence (K/L) grade 2 or worse. The case-control pairs were individually matched by sex age body mass index and race. For each hand x-ray 20 × 20 pixels bone texture regions were selected on the distal and middle phalanges adjacent to the DIP5 joint (Figure 1). For each bone region the FSSta StrS StdS parameters were calculated at scales of 0.34 and 0.51 mm. One-way analysis of variance ANOVA with Tuckey HSD (Games-Howell if appropriate) post hoc tests and paired samples t-tests (Wilcoxon signed-rank tests if appropriate) were used (p < 0.05 is significant).Results: For all image sizes and scales values of FSSta were statistically significantly different between isotropic fractal images. StrS values obtained for anisotropic surfaces were lower than those for isotropic surfaces and StdS agreed with the dominating directions. Compared to the controls OA middle phalanges exhibited significantly lower FSSta at sizes 0.34 and (p = 0.018) and 0.51 mm (p = 0.021) and higher StrS at 0.34 (p = 0.015) and 0.51 mm (p = 0.002). In the distal phalanx FSSta at size of 0.34 mm (p = 0.044) was lower for OA cases than controls. Conclusions: The AVOT method can differentiate between small isotropic and anisotropic fractal textures and also between finger bones with and without radiographic OA. Although further large-scale studies are still required our results show the potential of the method for the quantification of OA changes in the finger bone structure.