Quantification of differences in bone texture from plain radiographs in knees with and without osteoarthritis

Abstract

Purpose: Texture analysis of bone is not as dependent on the imaging conditions as direct evaluation of grayscale values from radiographs. The aim of the study was to investigate differences in bone texture between subjects with different stages of knee osteoarthritis and age- and gender-matched controls from plain radiographs using advanced image analysis methods. Methods: Standard anterior-posterior weight bearing radiographs (tube voltage = 60 kV, quantity of charge = 25 mAs) from 203 knees (103 male subjects, age: 50–69 years) were analyzed using MATLAB software (MathWorks inc., USA) and graded according to the Kellgren-Lawrence (KL) grading scale, in which 0 is normal and 4 is severe OA. KL grades were not known during the quantitative image analyses. Joint space width (JSW) was measured from the center of medial and lateral condyles. Four rectangle-shaped regions-of-interest (ROI) were extracted from the tibia and one elliptical-shaped ROI from the soft tissue beside the joint (Fig. 1). To assess bone density, mean grayscale value from the unprocessed ROIs (=GV) and mean grayscale value of the soft tissue ROI subtracted from GV (=GV’) were calculated. Local binary patterns (LBP) and second order partial derivatives (Laplacians) were calculated from the unprocessed ROIs to construct LBP and Laplacian-based images. In the LBP method, the 8 neighbor pixels for each pixel in the ROI were examined and an 8-bit LBP-value was calculated. From the LBP- and Laplacian-based images, Homogeneity index (HI) and entropy (E) texture parameters were calculated to evaluate bone structure. HI was derived from gray-level co-occurrence matrix. Some ROIs had to be excluded due to distractions (e.g. bright edges) from a piece of the clothing or similar artifact. If one-way analysis of variance (ANOVA) was statistically significant, Fisher’s least significant difference post-hoc test was performed to find out the KL groups that differed statistically significantly from each other. To evaluate intra-rater reproducibility of the analysis method, one investigator performed the analyses for a sub-population of 70 knees three times with two weeks interval. To evaluate inter-rater reproducibility, three investigators performed the analyses once for the same sub-population. Reproducibilities were evaluated using root-mean-square average coefficient of variation (CVRMS). Statistical analysis was performed using SPSS 19 software (SPSS Inc., USA). Results: Distribution of KL grades was the following: KL0 = 110 knees, KL1 = 28, KL2 = 27, KL3 = 31, and KL4 = 7. The intra-rater and inter-rater reproducibilities of the texture parameters were better (Range of intra-rater and inter-rater CVRMS values: 0.23–2.48%) than the bone density-related parameters (1.45–22.19%). Particularly, ELBP, ELap, and HILBP were highly reproducible (0.23–1.59%). Medial JSW was significantly (ANOVA: p < 0.05) higher in the control group (KL0) than in the KL2-4 groups. In the Fig. 2, differences between KL groups using GV, ELBP, and ELap are shown. Furthermore, GV’ and HILBP were significantly (p < 0.05) lower in the KL0 than in the KL2-4 groups in the medial subchondral bone plate. Similarly, HILBP was significantly (p < 0.05) lower in the KL0 than in the KL1-4 groups in the medial and lateral trabecular bone whereas GV’ was significantly (p < 0.05) lower in the KL0 than in the KL2-4 groups in the medial trabecular bone. Conclusions: Our results indicate that the changes in bone texture in knee OA can be quantitatively evaluated from plain radiographs using advanced image analysis methods. Based on the results, increased bone density, due to subchondral bone sclerosis, can be directly estimated from the grayscale values, if the X-ray imaging conditions are constant between patients. However, structural analysis of bone was more reproducible than direct evaluation of grayscale values, and it is therefore better suited for quantitative analysis when X-ray imaging conditions are variable.Fig. 2Statistically significant differences between KL groups in medial and lateral tibial subchondral bone plate (SBP) and trabecular bone (TB) using mean grayscale value (GV) and entropy from both Laplacian- based (ELap) and local binary patterns images (ELBP). * = Studied KL group differs significantly from the indicated KL group (LSD post-hoc test).View Large Image Figure ViewerDownload Hi-res image Download (PPT)