Predicting 3D pose in partially overlapped X-ray images of knee prostheses using model-based Roentgen stereophotogrammetric analysis (RSA)

Chi Pin Hsu, Shang Chih Lin, Kao Shang Shih, Chang Hung Huang, Chian Her Lee

Research output: Contribution to journalArticle

2 Citations (Scopus)

Abstract

After total knee replacement, the model-based Roentgen stereophotogrammetric analysis (RSA) technique has been used to monitor the status of prosthetic wear, misalignment, and even failure. However, the overlap of the prosthetic outlines inevitably increases errors in the estimation of prosthetic poses due to the limited amount of available outlines. In the literature, quite a few studies have investigated the problems induced by the overlapped outlines, and manual adjustment is still the mainstream. This study proposes two methods to automate the image processing of overlapped outlines prior to the pose registration of prosthetic models. The outline-separated method defines the intersected points and segments the overlapped outlines. The feature-recognized method uses the point and line features of the remaining outlines to initiate registration. Overlap percentage is defined as the ratio of overlapped to non-overlapped outlines. The simulated images with five overlapping percentages are used to evaluate the robustness and accuracy of the proposed methods. Compared with non-overlapped images, overlapped images reduce the number of outlines available for model-based RSA calculation. The maximum and root mean square errors for a prosthetic outline are 0.35 and 0.04 mm, respectively. The mean translation and rotation errors are 0.11 mm and 0.18°, respectively. The errors of the model-based RSA results are increased when the overlap percentage is beyond about 9 %. In conclusion, both outline-separated and feature-recognized methods can be seamlessly integrated to automate the calculation of rough registration. This can significantly increase the clinical practicability of the model-based RSA technique.

Original languageEnglish
Pages (from-to)1061-1071
Number of pages11
JournalMedical and Biological Engineering and Computing
Volume52
Issue number12
DOIs
Publication statusPublished - 2014

Fingerprint

Knee prostheses
Knee Prosthesis
Prosthetics
X-Rays
X rays
Knee Replacement Arthroplasties
Mean square error
Image processing
Wear of materials

Keywords

  • Knee
  • Overlap
  • RSA
  • TKR
  • X-ray

ASJC Scopus subject areas

  • Biomedical Engineering
  • Computer Science Applications
  • Medicine(all)

Cite this

Predicting 3D pose in partially overlapped X-ray images of knee prostheses using model-based Roentgen stereophotogrammetric analysis (RSA). / Hsu, Chi Pin; Lin, Shang Chih; Shih, Kao Shang; Huang, Chang Hung; Lee, Chian Her.

In: Medical and Biological Engineering and Computing, Vol. 52, No. 12, 2014, p. 1061-1071.

Research output: Contribution to journalArticle

@article{d9ca7499e8054d6794971d33d778628c,
title = "Predicting 3D pose in partially overlapped X-ray images of knee prostheses using model-based Roentgen stereophotogrammetric analysis (RSA)",
abstract = "After total knee replacement, the model-based Roentgen stereophotogrammetric analysis (RSA) technique has been used to monitor the status of prosthetic wear, misalignment, and even failure. However, the overlap of the prosthetic outlines inevitably increases errors in the estimation of prosthetic poses due to the limited amount of available outlines. In the literature, quite a few studies have investigated the problems induced by the overlapped outlines, and manual adjustment is still the mainstream. This study proposes two methods to automate the image processing of overlapped outlines prior to the pose registration of prosthetic models. The outline-separated method defines the intersected points and segments the overlapped outlines. The feature-recognized method uses the point and line features of the remaining outlines to initiate registration. Overlap percentage is defined as the ratio of overlapped to non-overlapped outlines. The simulated images with five overlapping percentages are used to evaluate the robustness and accuracy of the proposed methods. Compared with non-overlapped images, overlapped images reduce the number of outlines available for model-based RSA calculation. The maximum and root mean square errors for a prosthetic outline are 0.35 and 0.04 mm, respectively. The mean translation and rotation errors are 0.11 mm and 0.18°, respectively. The errors of the model-based RSA results are increased when the overlap percentage is beyond about 9 {\%}. In conclusion, both outline-separated and feature-recognized methods can be seamlessly integrated to automate the calculation of rough registration. This can significantly increase the clinical practicability of the model-based RSA technique.",
keywords = "Knee, Overlap, RSA, TKR, X-ray",
author = "Hsu, {Chi Pin} and Lin, {Shang Chih} and Shih, {Kao Shang} and Huang, {Chang Hung} and Lee, {Chian Her}",
year = "2014",
doi = "10.1007/s11517-014-1206-2",
language = "English",
volume = "52",
pages = "1061--1071",
journal = "Medical and Biological Engineering and Computing",
issn = "0140-0118",
publisher = "Peter Peregrinus Ltd",
number = "12",

}

TY - JOUR

T1 - Predicting 3D pose in partially overlapped X-ray images of knee prostheses using model-based Roentgen stereophotogrammetric analysis (RSA)

AU - Hsu, Chi Pin

AU - Lin, Shang Chih

AU - Shih, Kao Shang

AU - Huang, Chang Hung

AU - Lee, Chian Her

PY - 2014

Y1 - 2014

N2 - After total knee replacement, the model-based Roentgen stereophotogrammetric analysis (RSA) technique has been used to monitor the status of prosthetic wear, misalignment, and even failure. However, the overlap of the prosthetic outlines inevitably increases errors in the estimation of prosthetic poses due to the limited amount of available outlines. In the literature, quite a few studies have investigated the problems induced by the overlapped outlines, and manual adjustment is still the mainstream. This study proposes two methods to automate the image processing of overlapped outlines prior to the pose registration of prosthetic models. The outline-separated method defines the intersected points and segments the overlapped outlines. The feature-recognized method uses the point and line features of the remaining outlines to initiate registration. Overlap percentage is defined as the ratio of overlapped to non-overlapped outlines. The simulated images with five overlapping percentages are used to evaluate the robustness and accuracy of the proposed methods. Compared with non-overlapped images, overlapped images reduce the number of outlines available for model-based RSA calculation. The maximum and root mean square errors for a prosthetic outline are 0.35 and 0.04 mm, respectively. The mean translation and rotation errors are 0.11 mm and 0.18°, respectively. The errors of the model-based RSA results are increased when the overlap percentage is beyond about 9 %. In conclusion, both outline-separated and feature-recognized methods can be seamlessly integrated to automate the calculation of rough registration. This can significantly increase the clinical practicability of the model-based RSA technique.

AB - After total knee replacement, the model-based Roentgen stereophotogrammetric analysis (RSA) technique has been used to monitor the status of prosthetic wear, misalignment, and even failure. However, the overlap of the prosthetic outlines inevitably increases errors in the estimation of prosthetic poses due to the limited amount of available outlines. In the literature, quite a few studies have investigated the problems induced by the overlapped outlines, and manual adjustment is still the mainstream. This study proposes two methods to automate the image processing of overlapped outlines prior to the pose registration of prosthetic models. The outline-separated method defines the intersected points and segments the overlapped outlines. The feature-recognized method uses the point and line features of the remaining outlines to initiate registration. Overlap percentage is defined as the ratio of overlapped to non-overlapped outlines. The simulated images with five overlapping percentages are used to evaluate the robustness and accuracy of the proposed methods. Compared with non-overlapped images, overlapped images reduce the number of outlines available for model-based RSA calculation. The maximum and root mean square errors for a prosthetic outline are 0.35 and 0.04 mm, respectively. The mean translation and rotation errors are 0.11 mm and 0.18°, respectively. The errors of the model-based RSA results are increased when the overlap percentage is beyond about 9 %. In conclusion, both outline-separated and feature-recognized methods can be seamlessly integrated to automate the calculation of rough registration. This can significantly increase the clinical practicability of the model-based RSA technique.

KW - Knee

KW - Overlap

KW - RSA

KW - TKR

KW - X-ray

UR - http://www.scopus.com/inward/record.url?scp=84925541760&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84925541760&partnerID=8YFLogxK

U2 - 10.1007/s11517-014-1206-2

DO - 10.1007/s11517-014-1206-2

M3 - Article

VL - 52

SP - 1061

EP - 1071

JO - Medical and Biological Engineering and Computing

JF - Medical and Biological Engineering and Computing

SN - 0140-0118

IS - 12

ER -