Technical note: CT thresholding protocols for taking measurements on three‐dimensional models

Abstract

Computed tomography (CT) is a powerful analytical tool that is becoming increasingly popular for investigating small and difficult to visualize (internal) structures. However, along with this capability come numerous potential difficulties. One vexing problem is how to interpret the images that result from CT scanning. A typical CT slice displays a spectrum of gray-scale values that represent the densities of the structures in the image. To measure a particular structure it is necessary to establish a threshold value (i.e., gray-scale value), which clearly distinguishes the structure of interest from surrounding structures. However, determining the proper threshold is not a trivial task. One problem arises from the fact that the boundaries between adjacent structures (e.g., bone and air) are never clearly defined, but are instead represented by a continuum of gray-scale values. Another problem is related to the viewer control settings (window and leveling values), which can greatly affect the visual appearance of images, particularly along the edges of structures. Simply basing of the threshold value on the apparent (visual) boundaries has been shown to yield erroneous results (Koehler et al., 1979; Baxter and Sorenson 1981; Hara et al., 2002). A handful of studies have sought to investigate this problem, and these have led to several proposals regarding the appropriate method for establishing the threshold value. One approach that has enjoyed increasing attention is the half-maximum height protocol (HMH), which calculates the threshold value as the mean of the maximum and minimum gray scale values along a row of pixels that spans the boundary transition (Ulrich et al., 1980). This method has been found to produce accurate measurements of modern human vertebrae (using regular CT images—Ullrich et al., 1980; Baxter and Sorenson 1981; Seibert et al., 1981; Eubanks et al., 1985) and of extant and fossilized long-bones and teeth (Spoor et al., 1993). More recently, Fajardo et al. (2002) used a variation of the HMH method (applied to high resolution X-ray computed tomography) to accurately reconstruct and measure trabecular architecture of long bones. These researchers also found that it was important to sample the appropriate region of interest because bone types of different density (e.g., cortical versus trabecular) yield different HMH values (Fajardo et al., 2002). One of the limitations of these studies is that the HMH protocol and validation measurements were applied to individual slices. However, when a researcher wishes to take measurements on three-dimensional models created from CT, a single threshold value must be applied to the slices of interest. One potential concern is that applying a single threshold value to a dataset of dozens or hundreds of slices may not accurately represent the real morphology of various structures. Thus, this study sought to investigate the effects of taking linear measurements on three-dimensional models using a modified version of the HMH protocol. To examine the accuracy of the thresholding protocol, measurements were taken from high resolution X-ray computed tomography data and compared with measurements taken on dried specimens. The oval window was selected as the variable of interest because it represents a relatively small structure and also because it is possible to obtain these measurements on the dried skulls of certain taxa with a reasonable amount of accuracy. A Zeiss Discovery V.12 stereo digital microscope was used to take the measurements on the dried specimens (hereafter referred to as Zeiss-based measurements). All scanning was done at the University of Texas High-Resolution X-ray CT facility. These scans were taken with a 68-lm slice thickness and 68-lm interslice spacing. The images were reconstructed from 1,000 views and the field of view was 64 mm yielding a pixel size of 62.5 lm (1024 3 1024 pixel matrix). The final images were 16-bit TIFF files. The image stacks were imported into ImageJ 1.35f (NIH), cropped, converted to 16-bit signed files (to be compatible with the measurement software), and saved as raw stacks. These stacks were then used to construct three-dimensional models using 3D Slicer 2.6 and the measurements were taken using the fiducials module. This module allows the researcher to place markers (fiducials) anywhere on the model and take linear measurements between the markers. It is useful since it permits structures to be measured that are not in the same plane. Each measurement was taken three times and the mean of the three measurements was taken as the final value.