Elsevier

Pattern Recognition Letters

Volume 23, Issue 4, February 2002, Pages 395-403
Pattern Recognition Letters

Suitability of texture features to assess changes in trabecular bone architecture

https://doi.org/10.1016/S0167-8655(01)00172-6Get rights and content

Abstract

The purpose of this study was to determine the ability of texture features to assess changes in trabecular bone architecture as projected in radiographs. Micro-CT datasets of trabecular bone were processed to simulate different changes in architecture. Radiographs were simulated by projecting the 3D-bone structure. Texture features, based on mathematical morphology, determined on the simulated radiographs were able to detect structural changes in the trabecular bone.

Introduction

The strength of trabecular bone is determined by both the structure and the density of bone. In osteoporotic bone, the strength of trabecular bone is diminished, resulting in an increased fracture risk. For years, bone density has been studied as a major determinant of osteoporosis. This can be attributed to the fact that structure constitutes a feature, that is, compared to density, difficult to quantify non-invasively. Several studies, using CT or MRI, have been carried out to measure the structure of trabecular bone in vivo (Majumdar et al., 1998). The resolution, however, is still a limiting factor. As a cheaper alternative, plain radiographs have been used to quantify bone structure using different texture analysis methods (Luo et al., 1999; Pothuaud et al., 2000). However, a problem that is often overlooked arises: how to make sure that the texture features really quantify trabecular structure and are not merely density detectors? Ensuring that the texture features used are independent of first-order gray level statistics in the image, is a first step but not the solution for the problem. The real problem lies beneath the image plane in the 3D-trabecular structure: structure and density are two concepts that are physically closely intertwined and therefore strongly correlated. Each structure feature will show a correlation with density. When structure or texture features are used in a classification experiment, the question is which part of the discriminative power can be attributed to bone density, and which part to structural differences. By using 3D-datasets of trabecular bone, changes of structure and density can be simulated (semi-) independently. By projecting the trabecular structure onto a plane, a radiograph can be simulated. Using these radiographs, the ability of texture features to assess changes in bone structure can be determined.

However, most texture analysis methods have not been specifically developed for use on radiographs. Radiographs differ essentially from other imagery, in the sense that they are dose limited: for patient safety, the exposure dose is kept as low as possible. In order to enhance the light output, intensifying screens are used in front of the film. These screens have a blurring effect. In the spatial frequency domain, the modulation transfer function (MTF) describes this effect. This blurring effect results in a reduced spatial resolution, contrast, and sharpness. It has been shown that texture features are sensitive to this blurring effect (Veenland et al., 1998). Therefore, the ability of texture features to quantify changes in structure should be evaluated taking into account the influence of the MTF on texture.

The goal of this study was to investigate whether texture features, determined in radiographs of trabecular bone, are able to quantify changes in the underlying 3D-trabecular bone network. In order to determine whether texture features contain real structure information, which has an added value to density, two models for osteoporosis were used. In the “erosion model”, both structure and density were diminished. In the “opening model” the bone mass was kept more or less constant, while the architecture of the bone was changed: thin trabeculae were perforated, while the thickness of thicker trabeculae was maintained. The models were based on micro-CT datasets of trabecular bone. We simulated radiographs by projecting the trabecular structures onto a 2D-plane. On these projections the blurring effect of different screen-film combinations were simulated. The ability of texture features, determined on these projections, to discriminate between different stages of the erosion and the opening model were quantified. We compared the performance of texture features to bone density and investigated the influence of MTFs on the discriminative performance.

Section snippets

Material and methods

In order to study the suitability of texture features to assess osteoporotic changes in projected textures, two computer models were used. Both models were based on 3D-datasets of trabecular bone, imaged by micro-CT. In the first model, the process of osteoporosis was simulated by simply eroding the trabecular bone in consecutive steps. By this erosion process, both the structure and the amount of bone changed. In the second model, the bone structure was changed while preserving the bone mass:

Results

In Table 2, the features that were selected by the feature selection method are listed. Different features were selected for the different models. Furthermore, for the erosion model, using the unfiltered projection or the projections on which the effect of the two different MTFs were simulated, resulted in the choice of other features. For the opening model, the choice of features was not influenced by the MTF that simulates the effect of a sharp screen-film combination (MTF 1). Different

Conclusions

In this paper, we investigated whether texture features, determined in radiographs of trabecular bone, are able to quantify changes in the underlying 3D-trabecular bone network. For this purpose we used two models based on 3D-micro-CT datasets of trabecular bone: an erosion model and an opening model. In the erosion model, both structure and density were diminished. In the opening model the bone mass was kept more or less constant, while the architecture of the bone was changed: thin trabeculae

References (9)

There are more references available in the full text version of this article.

Cited by (27)

  • Classification of the trabecular bone structure of osteoporotic patients using machine vision

    2017, Computers in Biology and Medicine
    Citation Excerpt :

    Conventional bone radiographs are considered as a cheaper alternative to DXA. Moreover, it has been shown that texture analysis of X-ray radiographs is well suited to assess changes in trabecular bone architecture [41,42]. The aim of the approach proposed here is to complete the diagnosis achieved in clinical routine based on the BMD measured by DXA, by characterizing and quantifying the quality of the bone micro-architecture.

  • Combination of radiograph-based trabecular and geometrical parameters can discriminate cervical hip fractures from controls in individuals with BMD in non-osteoporotic range

    2011, Bone
    Citation Excerpt :

    In order to achieve a more comprehensive assessment of fracture risk, recent efforts have been directed on different imaging modalities, including high resolution computed tomography (CT) and magnetic resonance imaging (MRI) [16–21]. In addition to these three-dimensional imaging modalities, analysis of plain radiographs has also been proposed with promising results [22–31]. Since the method is widely available with low imaging costs, it has been of considerable interest to discover how well it could be utilized for the assessment of fracture risk.

  • Morphological texture description of grey-scale and color images

    2011, Advances in Imaging and Electron Physics
    Citation Excerpt :

    While Sariyanni et al. (2001) have performed a quantitative study for the discrimination of different hepatic lesions using morphological and other fractal dimensions. Veenland et al. (2002), on the other hand, have studied the ability of texture features to assess changes in trabecular bone architecture as projected in radiographs. Specifically, they have concentrated on the ratios of the differences of morphological gradients with various SE shapes that provide invariance against multiplicative intensity changes.

  • The use of visual assessment of dental radiographs for identifying women at risk of having osteoporosis: the OSTEODENT project

    2008, Oral Surgery, Oral Medicine, Oral Pathology, Oral Radiology and Endodontology
    Citation Excerpt :

    Risk factors for low BMD have been shown to be the same as risk factors for fracture12 and the method described in this study could be one way of identifying women with or without risk factors for low BMD. Other bone tissue properties than mineral density contribute to strength and, although osteoporosis is still defined in terms of bone density, there is supporting literature that structural or architectural changes of the trabecular bone are as important.13,14 One way of imaging structural change is plain radiography, and several texture analysis methods have been used for this purpose.15,16

View all citing articles on Scopus
View full text