gms | German Medical Science

Deutscher Kongress für Orthopädie und Unfallchirurgie
74. Jahrestagung der Deutschen Gesellschaft für Unfallchirurgie
96. Tagung der Deutschen Gesellschaft für Orthopädie und Orthopädische Chirurgie
51. Tagung des Berufsverbandes der Fachärzte für Orthopädie und Unfallchirurgie

26. - 29.10.2010, Berlin

Quantification of the Accuracy of MRI Generated 3D Models of Long Bones

Meeting Abstract

  • K. Rathanayaka - Queensland University of Technology, Institute of Health and Biomedical Innovation, Brisbane, Australia
  • K. Momot - Queensland University of Technology, School of Physical and Chemical Sciences, Brisbane, Australia
  • H. Noser - AO Research Institute, Human Morphology Service Center, Davos Platz, Switzerland
  • T. Sahama - Queensland University of Technology, School of Information Technology, Brisbane, Australia
  • M.A. Schuetz - Queensland University of Technology, Institute of Health and Biomedical Innovation, Brisbane, Australia
  • B. Schmutz - Queensland University of Technology, Institute of Health and Biomedical Innovation, Brisbane, Australia

Deutscher Kongress für Orthopädie und Unfallchirurgie. 74. Jahrestagung der Deutschen Gesellschaft für Unfallchirurgie, 96. Tagung der Deutschen Gesellschaft für Orthopädie und Orthopädische Chirurgie, 51. Tagung des Berufsverbandes der Fachärzte für Orthopädie. Berlin, 26.-29.10.2010. Düsseldorf: German Medical Science GMS Publishing House; 2010. DocIN16-1086

DOI: 10.3205/10dkou092, URN: urn:nbn:de:0183-10dkou0927

Veröffentlicht: 21. Oktober 2010

© 2010 Rathanayaka et al.
Dieser Artikel ist ein Open Access-Artikel und steht unter den Creative Commons Lizenzbedingungen (http://creativecommons.org/licenses/by-nc-nd/3.0/deed.de). Er darf vervielfältigt, verbreitet und öffentlich zugänglich gemacht werden, vorausgesetzt dass Autor und Quelle genannt werden.


Gliederung

Text

Objective: Orthopaedic implants are typically designed using accurate computer tomography (CT) based 3D models of long bones. CT scans are used due to their high soft tissue-bone contrast. However, CT scanning exposes a subject to a high amount of ionising radiation making it unsuitable for acquisition of data from healthy human volunteers for research purposes. The alternative use of cadaver specimens is also limited, as most of the available cadavers are from old donors. Magnetic resonance imaging (MRI) is not routinely used for imaging of bones due to the difficulty of segmentation between bone and certain soft tissue types. However, MRI has the advantage of not using ionising radiation and could be a suitable alternative to CT. Therefore, this study aims to quantify the accuracy of MRI based 3D models of long bones compared to CT based 3D models.

Methods: CT scans of five ovine femora were acquired with a resolution of 0.4 mm x 0.4 mm × 0.5 mm. MRI data of the same specimens was acquired using a clinical 1.5T scanner with a resolution of 0.45 mm × 0.45 mm × 1 mm. For both MRI and CT, intensity thresholding was performed using three different threshold levels for the proximal, shaft and distal regions. To minimise user-dependent errors, appropriate threshold levels were determined with the aid of a method based on the Canny edge detection filter. The segmentation and reconstruction of the 3D models was performed with Amira. The reference 3D models were generated by digitising the bone surface with a contact 3D scanner (Roland MDX-20, resolution: 0.3 mm x 0.3 mm x 0.025 mm). The geometric differences between a model of interest and the reference model were calculated as average distance by utilising the reverse engineering software Rapidform. Differences were also calculated for the different anatomical regions; the head, proximal, shaft, distal and distal articular regions, to localise the errors associated with different regions of the bone.

Results and conclusions: The surface geometry of the CT derived models displayed a high accuracy with a mean error of 0.15 mm. In comparison, the MRI-derived models contained slightly higher errors presenting a mean error of 0.23 mm, when validated with the reference models. From the different anatomical regions, the shaft presented the lowest errors of 0.07 mm for CT and 0.15 mm for MRI. The larger MRI slice spacing (1 mm) compared to CT (0.5 mm) is most likely to have an effect on the surface geometry of the reconstructed model while poor segmentation in MRI also has a contribution, especially in the distal and proximal regions of the bone.

This study reveals that the accuracy of the MRI derived long bone models is comparable to the CT derived models, indicating that MRI data can be acquired from the target patient population. This enables designing anatomically better fitting orthopaedic implants, which in turn will improve outcomes for patients.