1. Intensity-based biplane 2D-3D registration for in-vivo three dimensional forearm rotational analysis
Shingo Abe1,3, Kunihiro Oka1, Yoshihito Otake2, Yuusuke Tennma2, Yuuta Hiasa2, Yoshinobu Sato2, Satoshi Miyamura1, Atsuo Shigi1, Tsuyoshi Murase1
1Department of Orthopaedic Surgery, Osaka University Graduate School of Medicine, Suita Japan
2Nara Institute of Science and Technology, Ikoma, Japan
3Toyonaka Municipal Hospital, Toyonaka, Japan
Full supination
Neutral
Full pronation
Virtual motion
【Introduction】
We have conventionally performed forearm computed tomography (CT) in three rotational position (full supination/ pronation, and neutral rotation) (Figure 1) and created virtual forearm rotation (Video 1) using the computer software (CT method). The purpose of this study was to create a new intensity-based 2D-3D registration (2D-3D method) to measure real forearm rotational motion and validate the accuracy of this method.
Video 1. Virtual forearm rotational motion using conventional CT method
Figure 1. CT at three positions
【Methods】
2D-3D registration algorithm employed in the study was summarized in Figure 2. Biplane fluoroscopy was recorded for 10 seconds during forearm rotation with 12.5 fps and a computed tomography (CT) was acquired at one static position. Digitally reconstructed radiograph (DRR), which was a projected image of the 3D CT data containing intensity information, was created for radius and ulna, respectively. The DRR of the radius and ulna was registered on the corresponding bone on fluoroscopic images for each frame based on a similarity metric which calculated intensity value of pixels. All the registration procedure was implemented through an automatic matching algorithm written by MATLAB software.Consequently, 6 degree-of-freedom rotation and translation for the radius and the ulna during forearm rotation were calculated at each frame (125 frames in total).
For validation study, we manufactured the upper arm phantom (Figure 3, Kyoto Kagaku, Japan) composed of bones, soft-tissue, and 8 stainless spheres with diameter 1.5mm implanted on the radius and the ulna. We defined the radiostereometric analysis (RSA method), which calculated 3D bone position based on 3D position of stainless spheres, as ground truth. Because the density of spheres on the images help to increase the accuracy of registration, we removed the density of spheres from the images using the impainting technique (Figure 4) which replace the density of the spheres to average density of surroundings. We executed 2D-3D method using the images without the stainless spheres. To assess accuracy of 2D3D method, we compared 6 DOF of rotation and translation for the radius and ulna between 2D-3D method and RSA method. The coordinate system for the radius and ulna were shown in Figure 5.
Figure 4. (A) The original image of one of the frames of fluorography and CT. (B) The impainting technique was done to replace the intensity of the stainless spheres
Figure 2. The algorithm of 3D/2D registration technique are shown.
Figure 3. Upper arm phantom
【Results】
The absolute value of rotation error were 0.36 ± 0.59° for the radius, and 0.55 ± 0.73° for the ulna. The absolute value of translation error were 0.26 ± 0.28 mm for the radius, and 0.18 ± 0.18 mm for the ulna. A representative case of the proposed in-vivo forearm rotation analysis in Video 2.
Radiation dose was 1.58 mGy for 2D-3D method, which was lower than CT method (4.05 mGy).
Video 2. Forearm rotation using intensity based biplane 2D/3D registration
Figure 5. The coordinate system of the radius and ulna.
【Conclusion】
We achieved highly accurate in-vivo forearm rotational analysis using intensity-based biplane 2D-3D registration technique with high temporal and special resolution. In addition, the radiation dose of the current 2D-3D method was lower than conventional CT method.