THE MEASUREMENT OF BONE QUALITY USING GRAY LEVEL CO-OCCURRENCE MATRIX TEXTURAL FEATURES

Abstract

In this paper, statistical methods for the estimation of bone quality to predict the risk of fracture are reported. Bone mineral density and bone architecture properties are the main contributors of bone quality. Dual-energy X-ray Absorptiometry (DXA) is the traditional clinical measurement technique for bone mineral density, but does not include architectural information to enhance the prediction of bone fragility. Other modalities are not practical due to cost and access considerations. This study investigates statistical parameters based on the Gray Level Co-occurrence Matrix (GLCM) extracted from two-dimensional projection images and explores links with architectural properties and bone mechanics. Data analysis was conducted on Micro-CT images of 13 trabecular bones (with an in-plane spatial resolution of about 50μm). Ground truth data for bone volume fraction (BV/TV), bone strength and modulus were available based on complex 3D analysis and mechanical tests. Correlation between the statistical parameters and biomechanical test results was studied using regression analysis. The results showed Cluster-Shade was strongly correlated with the microarchitecture of the trabecular bone and related to mechanical properties. Once the principle thesis of utilizing second-order statistics is established, it can be extended to other modalities, providing cost and convenience advantages for patients and doctors.

Keywords: DXA; GLCM; bone; medical imaging; osteoporosis; texture.

Fig. 1

Transformation of 3D image into a 2D projection image

Fig. 2

(a) Gray scale image of 2D projection bone (b) template binary image with “0” inside interest region (c) binary template image after filling process, white stands for valid bone region

Fig. 3

(a) Original bone image (b) histogram equalization for the whole image (c) histogram equalization only on the valid bone part

Fig. 4

Linear regression models for bone parameter Elastic Modulus and (a) Energy R² = 0.395 and (b) Homogeneity R² = 0.401.

Fig 5

R² values for bone parameters with respect to GLCM features when d=1 and the four directional results are averaged

Fig. 6

“Coefficient of Determination R²” for bone volume fraction BV/TV with respect to GLCM measure Cluster Shade when d=4 and orientation = 0°

Fig. 7

GLC Matrix for bone with low Strength for 0.9 Mpa (displacement for 4, quantization levels 8, orientation 90 degrees).