Effects of model composition and number of image sources on the accuracy of model-based 3D/2D image registration methods for measuring three-dimensional knee kinematics

• Published in: Journal of Biomechanical Science and Engineering Vol. 16; no. 3; p. 21-00105
• Main Authors: LIN, Cheng Chung, LU, Hsuan Yu, LU, Tung Wu
• Format: Journal Article
• Language: English
• Published: Tokyo The Japan Society of Mechanical Engineers 2021; Japan Science and Technology Agency
• Subjects: 3D fluoroscopy; Accuracy; Cadavers; Composition effects; Computed tomography; Configurations; Density; Evaluation; Fluoroscopic imaging; Image reconstruction; Image registration; Joints (anatomy); Kinematics; Knee; Measurement methods; Medical imaging; Model accuracy; Multiple x-ray images; Radiodensity; Registration; Skeleton; Three dimensional models; Two dimensional models
• ISSN: 1880-9863; 1880-9863
• DOI: 10.1299/jbse.21-00105

Model-based three-dimensional(3D)/two-dimensional(2D) image registration methods have been widely applied in measuring 3D kinematics of the knee during dynamic activities. However, the combined effects of bone model compositions (radiodensity vs. homogeneous-density) and the number of fluoroscopic views on the measurement accuracy remained unclear. The current study evaluated experimentally the accuracy of the four model-based 3D/2D image registration configurations on the accuracy of measured knee kinematics, namely homogeneous-density model/single-plane image (HS), radiodensity model/single-plane image (RS), homogeneous-density model/biplane images (HB), and radiodensity model/biplane images (RB). Computed tomography (CT) of the knee and asynchronous biplane fluoroscopic images of the simulated knee motions were collected from a cadaveric knee joint for the evaluation of the registration configurations. The results showed that the use of biplane fluoroscopic images ensured mean absolute errors (MAE) below 0.3 mm and 0.9° in each motion component regardless of the types of bone models. Application of radiodensity model could generate digitally reconstructed radiographs more similar to the fluoroscopic images, diminishing MAE in all motion components and measurement bias. As a result, the RS configuration was capable of reconstructing the 3D knee joint angles with MAE comparable to those obtained using the HB configuration. Among the four tested configurations, the RB configuration was most accurate and least affected by the fast skeleton motions.