Computerized methods for X-ray-based small bone densitometry

https://doi.org/10.1016/S0169-2607(02)00164-5Get rights and content

Abstract

Animal models have been widely used to correlate in vivo changes in bone mineral density (BMD) with changes in disease state of bone. In small animal models, e.g. the hindlimb suspension model of bone loss, a non-invasive assessment of BMD is required. X-ray radiography has been surpassed in some cases by quantitative computed tomography (QCT) and dual X-ray absorptiometry (DEXA) quantitation. However, there are drawbacks in using the computerized methods, especially for small animals. In this paper, we present image-processing algorithms to quantitatively determine bone area and mineral density in digitized radiographs. Image calibration is based on a calibration step wedge, and the algorithm automatically detects the steps and computes the calibration data. In addition, we demonstrate how the algorithm can accurately determine the cortical outline of the bone and provide reliable data and statistics for small animal studies. A downloadable implementation example for the popular NIH Image package is provided.

Section snippets

Introduction and background

A major difficulty in evaluating in vivo changes in small animal studies has been the lack of accurate and precise methods to non-invasively determine bone quality. In studies of osteoporosis and osteopenia (bone loss) in human patients [1], [2], [3], [4], [5] and in animal models such as the hindlimb suspension model [6], [7], [8], the state of the bone is routinely assessed by measurement and interpretation of bone mineral density (BMD). BMD is the amount of bone mineral, calcium

Design considerations

We perform analysis of BMD routinely in our lab [21]. When attempting to use mice instead of rats, however, it became obvious that resolution limitations of DEXA would preclude us from using this method. In addition, initially the outline of the bone was traced manually, leading to long analysis times and potential subjective observer influences. For this reason, we developed a computer algorithm based on the methods described by Colbert et al. [17] and Colbert and Bachtell [22]. The main

Image acquisition and processing

All X-ray images were exposed on a HP 43805N X-ray machine using Kodak X-Omat TL film. The developed film was digitized at 500 dpi and 12 bits per pixel (bpp) using an Agfa Duoscan 2500 scanner. Image analysis was performed with Scion Image (Scion Corp., Frederick, MA), which is a PC port of NIH Image. Algorithms were designed using the NIH Image macro-language. A step wedge used as calibration phantom was custom-machined from aluminum with 0.5 mm step height increments.

Bone density and geometry phantoms

To characterize the

Status report

The implementation of the algorithms using NIH Image has been used in our lab for 2 years. Generally, it was found that both the wedge detection and the bone contour detection worked satisfactorily. Representative examples are given below.

Lessons learned

The algorithm has been routinely used for over 2 years. The main goals (shorter analysis time and reduced random scatter) have been achieved. Particularly effective was the introduction of the soft tissue correction, because the X-ray attenuation caused by soft tissue was in the same order of magnitude as the attenuation caused by the inner cavity of the femoral bone. In other words, bone density appeared elevated by up to 100%, the exact value partially depending on the thickness of the

Future plans

Since the algorithm has been proven to be robust and reliable, future plans include an implementation using the C programming language. First experiments using the GTK visual toolkit showed that the two disadvantages of the NIH Image implementation could be overcome. Bone segmentation requires about 100–200 ms in a compiled language, which allows real-time manipulation of the parameters. Also, array handling and post-processing of the segmentation results allow the detection of segmentation

Acknowledgements

We thank David Simonton at the Radiology Research Lab of the San Diego Veterans Administration Hospital for the preparation of the radiographs. This work was funded by NIH grants AR-46797 and HL-40696.

References (23)

  • T. Sakata et al.

    Trabecular bone turnover and bone marrow cell development in tail-suspended mice

    J. Bone Miner. Res.

    (1999)
  • Cited by (20)

    • Letrozole induced changes in bone mineral properties and mechanical functions of laying hens

      2019, Poultry Science
      Citation Excerpt :

      The original images were calibrated by the step-wedge image in terms of millimeters of aluminum equivalence. A macro-routine was written to delineate the boundary edge of each excised bone in order to calculate mean radiographic density for pixels within the boundary (Haidekker et al., 2004). The bone mineral density (BMD) was expressed in millimeters of aluminum thickness (mmAl).

    • Letrozole inhibits the osteogenesis of medullary bone in prelay pullets

      2010, Poultry Science
      Citation Excerpt :

      The NIH Image program displayed each image in false color and calibrated each x-ray from the step-wedge image in terms of millimeters of aluminum equivalence. A macro-routine was written to delineate the boundary edge of each excised bone and calculate mean radiographic density for pixels contained within the boundary (Haidekker et al., 2004). A 3-mm mid-diaphyseal scleromata was collected from each left tibia using a precision saw (Isomet 2000, Buehler, Lake Bluff, IL).

    View all citing articles on Scopus
    View full text