A Digital Model to Simulate Effects of Bone Architecture Variations on Texture at Spatial Resolutions of CT, HR-pQCT, and μCT Scanners

Abstract

The quantification of changes in the trabecular bone structure induced by musculoskeletal diseases like osteoarthritis, osteoporosis, rheumatoid arthritis, and others by means of a texture analysis is a valuable tool which is expected to improve the diagnosis and monitoring of a disease. The reaction of texture parameters on different alterations in the architecture of the fine trabecular network and inherent imaging factors such as spatial resolution or image noise has to be understood in detail to ensure an accurate and reliable determination of the current bone state. Therefore, a digital model for the quantitative analysis of cancellous bone structures was developed. Five parameters were used for texture analysis: entropy, global and local inhomogeneity, local anisotropy, and variogram slope. Various generic structural changes of cancellous bone were simulated for different spatial resolutions. Additionally, the dependence of the texture parameters on tissue mineralization and noise was investigated. The present work explains changes in texture parameter outcomes based on structural changes originating from structure modifications and reveals that a texture analysis could provide useful information for a trabecular bone analysis even at resolutions below the dimensions of single trabeculae.

Figure 1

Basic trabecular bone model. For clarity the 3D view only shows a detail of 5 × 5 × 5 rods.

Figure 2

Examples of modifications of the basic model. PLM: 60% of rods are deleted; BFM: rod diameter and plate thickness of 300 μm each; RLM: six plates remaining; SMM: surface irregularities with a width of 80 μm.

Figure 3

Structure modifications at a voxel size of 10 μm and a noise level of 30 HU. On the x-axis, the following measures are plotted. PLM (plate-like model): ratio of rods being deleted in %/100; BFM (bone formation model): BV/TV; RLM (rod-like model): number of plates remaining; SMM (surface modification model): width of surface irregularity in voxels. Results for the basic model are highlighted by circles. % changes between values at the basic model and maximal structure change are given as Δ.

Figure 4

Dependence of texture parameters on tissue mineralization for the basic model at a voxel size of 10 μm and a noise level of 30 HU. Percentage changes between values at 800 HU and 1200 HU are given as Δ.

Figure 5

Structure modifications at a voxel size of 90 μm and a noise level of 30 HU. On the x-axis, the following measures are plotted. PLM (plate-like model): ratio of rods being deleted; BFM (bone formation model): BV/TV; RLM (rod-like model): number of plates remaining; SMM (surface modification model): width of surface irregularity.

Figure 6

Structure modifications at a voxel size of 250 μm and a noise level of 30 HU. On the x-axis, the following measures are plotted. PLM (plate-like model): ratio of rods being deleted; BFM (bone formation model): BV/TV; RLM (rod-like model): number of plates remaining; SMM (surface modification model): width of surface irregularity.

Figure 7

Example of linear regression analysis. Local anisotropy at PLM (plate-like model) changes. Linear correlation between results from voxel sizes (250 μm) and reference voxel size (10 μm).

Figure 8

Dependence of texture parameters on image noise for the basic model at a voxel size of 10 μm. Percentage changes between values at noise = 5 HU and noise = 50 HU are given as Δ.