Knowing and understanding the correlation between size and density of detectable bone structures in clinical computed tomography (CT) is the key for many diagnosis purposes. To be able to distinguish at what sizes bone structures are detectable with a clinical CT scanner, parameters from the image model have to be compared to quantitative size parameters of the bone structures.
Hangartner et al. was able to show that bone width can be measured down to a width of less than 1 mm, by using algorithm based on iteratively varying CT thresholds for narrow structures and fixed thresholds for wide structures . However, they did not specify the bone density of their specimens, i.e. if it is relatively low or not.
CT is an excellent source of creating geometrically accurate 3D models by means of good segmentation methods and it is possible to convert the entire range of Hounsfield units to bone density . However it should be noted that images of very thin structures show artificially decreased density values due to the well known partial-volume effects .
In this work a standard clinical CT scanner is evaluated for detecting small bone structures in codfish and salmon. A 3D model of the fish skeletons are developed and few specific bones are analyzed to study differences in size parameters. This is done to see if parts of the bones are not detectable by the use of a CT scanner or not included in the model due to its small thickness or low density. It is hypothesized that parts of extremely thin bone structures or of low density will be absent in the 3D model. Helgason et al. did similar research on detecting thin bones and modeling for the development of automated bone detection methods for the food processing industry . Results are promising for future studies in this field.
2 Material and methods
Fish bones from a codfish and salmon were used as specimens. The two fishes were scanned separately in a standard clinical CT scanner, same parameters used for both scans. The CT data were then processed so that the skeleton was separated from the tissue and a 3D model developed. Quantitative measurements of selected bones were then compared to parameters observed from the model.
2.1 Codfish and salmon
The codfish was scanned the same day as it was caught off the coast of Iceland. We got the salmon from a fish farmer and it was kept in a freezer overnight before imaging. After imaging, the fishes were kept in a freezer until specific bones of interest were cut out for further analysis.
2.2 CT imaging system
The CT imaging system used was a scanner designed for human clinical use, Philips Brilliance 64. The scan parameters were set at the lowest energy of 80 KV with 0.67 mm slice thickness with 50% redundancy, i.e. each slice covers half of the volume as the previous slice covers. Field of view is 344 mm and each slice has an image matrix of 512 × 512 pixels.
2.3 Image processing – segmentation
Image processing was done using Mimics® (v15.0, Materialise NV, Leuven, Belgium) the specially developed software for medical image processing. Mimics® allows for easy and quick creation of accurate 3D models and segmentation from the CT imaging data.
CT image values are linearly transformed to Hounsfield units (HU) by relating the CT values to nearby water and air values calculated from the following equation .(1)
where µw and µA are the CT image values of water and air. Bone density is defined as mean value expressed in HU in each pixel  and in a 3D image, the picture elements are referred to as voxels, volume elements.
The CT slice images obtained along the length of the bodies of the fishes can be stacked to form a 3D representation of the bodies. Each voxel in the 3D image is characterized by its HU value and specific tissue types can be segmented from the rest of the body by sorting out the voxels with their values of HU in a detailed interval. Resulting in a geometrical 3D model of the specified tissue. The method of bone segmentation used in Mimics® is described briefly in the following steps:
– A mask is created by the means of thresholding, the first action performed to create a segmentation mask. Region of interest is selected by defining an interval of gray values, interval of 226 to 3071 HU was chosen, i.e. the mask consists only of voxels containing HU from this interval.
– Region growing is used to separate masks into different parts and eliminate noise by getting rid of floating pixels. The region-growing tool separates a structure from a mask by growing out from a starting voxel within the region of interest and finding the voxels that are connected to it.
– Additional voxels that have yet failed to connect are manually added to the mask.
– Erode region growing is used to eliminate voxels that do not belong to the anatomical structure of interest. The erode tool takes away the number of pixels selected. Then using the region-growing tool again creates a new mask.
– Dilate region growing is used if too many voxels were deleted from the mask. The dilate tool adds the number of pixels selected to the boundary of the mask. Performing the function erode followed by dilation is useful for breaking small connections. Here we have obtained the geometrical structure of interest, the skeleton of the fish.
– Boolean operations are used to separate the structure into smaller structures of interest.
2.4 Bone specimens
Small structures of interest were chosen from the models of the fish skeletons for further analysis. From the codfish two sets of bones from the headmost joints from the left side were chosen (see Figure 1). From the salmon a whole vertebra was chosen with six bones attached (see Figure 2).
The specific bone structures were harvested from the fish bodies. Disposable clinical scalpels were used for cutting. Afterwards the bones were soaked in 80 °C water to extract remaining flesh from the bones.
All measurements on the model were performed using measurement tools in Mimics®. Quantitative measurements of the selected bones consist of:
Mass: all bones were weighted on the same scale with d = 0.1 mg/mg.
Length: Optical measurements by photographing the bones lying on a millimeter paper for accurate scaling in ImageJ, imaging processing software (see Figure 3). Estimated accuracy of 0.25 mm. The model length measurements are estimated to have less accuracy due to the rough surface of the modelled bones making it difficult to measure in a straight line over the surface. Estimated accuracy of 2.50 mm.
Diameter: The diameters of the bones were measured with a micrometer screw gauge with a rated accuracy of 0.01 mm. The diameter was measured at three positions, at the bones fixation end, their middle and far end. Since the bones are not completely round the diameter was measured in two perpendicular directions, i.e. one in the direction of the length axis of the fish and other perpendicular to the length axis. It should be noted that the irregular shape of the bones slightly affects the accuracy of the diameter measurements.
Diameter: Comparisons of measured diameter with the millimeter screw to diameter measurements of the model are shown in the Table 3 and Table 4. The tables show results at three positions measured both in the direction of the length axis of the fish and perpendicular to the length axis.
4 Discussion and conclusion
The thin bone structures of the codfish and salmon are detected with a clinical CT and can be modeled. This is useful for detection of small bone fractures or thin bone structures.
It is interesting to see that the specimen bones of the codfish are thicker than of the salmon. The result show significantly larger difference in modeled parameters and measured ones for the salmon than the codfish specimen. Our results show ability to reconstruct closely the diameter of the codfish bones but to a less extent the small bones of the salmon.
Our results indicate that very thin bone structures are missing from the models due to the small thickness or low density resulting from the partial-volume-effect. Furthermore it is likely that edges of thin bone structures that are close together have merged and the structures have been modeled as a single piece. This can be improved by using a scanner with higher spatial resolution since it influences the representation of edges between objects of different densities, decreasing the field of view would also contribute to the improvement. Additionally the selection of appropriate threshold is essential to the accuracy of the geometrically segmented model.
In terms of accuracy of our models we can see that the codfish model is very close to measured values of the specimen bones (9.7 ± 9.3%), however the level of accuracy for the salmon model decreases significantly (38.5 ± 15.3%). Again this indicates that very thin bones structures or of very low densities are not completely detectable with a clinical CT. To further analyze our model it would be interesting to determine the levels of densities of the bones and compare it to the observed HU in the model. Further research on the subject is ongoing.
Special thanks to Haraldur Auđunsson for the codfish and to the company Rifós hf. for the salmon.
Hangartner TN., Short DF. Accurate Quantification of Width and Density of Bone Structures by Computed Tomography. In: Med Phys. 2007;34(10):3777-3784. Google Scholar
Rho JY, Hobatho MC, Ashman RB. Relations of Mechanical Properties to Density and CT Numbers in Human Bone. In: Med Eng Phys. 1995;17(5):347-355. Google Scholar
T. Helgason, R.A. Gudmundsdottir, K.L. Valtysdottir, K. Andersen. Detecting Thin Bones and Modeling Cod Skeleton. In: Proceedings of the European Modeling and Simulation Symposium, 2012; 163-168. Google Scholar
T. Helgason, P. Gargiulo, S. Knútsdóttir, V. Gudmundsdottir, H. Kern, U. Carraro, P. Ingvarsson, S. Yngvason, 2011. “Monitoring Muscle Growth and Tissue Changes Induced by Electrical Stimulation of Denervated Degenerated Muscles with CT and Stereolithographic 3D Modelling.” In: Narayan, R, Calvert, P, eds. Computer Aided Biomanufacturing. Weinheim Germany, Wiley-VCH, 130-146 Google Scholar
P. Gargiulo, U. Carraro, T. Mandl, H. Kern, S. Zampieri, W. Mayr, T. Helgason, 2012. Anthropometry of Human Muscle Using Segmentation Techniques and 3D Modelling: Applications to Lower Motor Neuron Denervated Muscle in Spinal Cord Injury. In: Preedy, V.R., ed. Handbook af Anthropometry. Physical Measures of Human Form in Health and Disease. US:Springer, 323-354 Google Scholar
Johannesdottir F., 2006. Bone: use it or lose it. MS thesis. University of Iceland. Google Scholar
About the article
Published Online: 2015-09-12
Published in Print: 2015-09-01
Conflict of interest: Authors state no conflict of interest. Material and Methods: Informed consent: Informed consent has been obtained from all individuals included in this study. Ethical approval: The research related to human use has been complied with all the relevant national regulations, institutional policies and in accordance the tenets of the Helsinki Declaration, and has been approved by the authors’ institutional review board or equivalent committee.