In this study, XPlane TEE images and reconstructed single plane TEE images of an ex vivo porcine heart were registered to a CT-derived surface of the aortic root. The results of the XPlane TEE to CT registration were combined with the TEE- fluoroscopy registration described in Chapter 4 to register the CT to fluoroscopy, with the transformations used shown in Figure 5.8, and registration accuracy was assessed.
5.3.1
Methods
5.3.1.1 Equipment and Materials
Theex vivo porcine heart containing 5 pin targets described in Section 4.6.1.4 was imaged using TEE, CT and fluoroscopy.
Fluoroscopy and CT images were acquired on a Medtronic O-arm Surgical Imag- ing SystemTM. The cone-beam CT was acquired at a voxel spacing of 0.41mm× 0.41mm×0.83mm, dimensions of 512×512×192 voxels, 120kVp, and 25mA. Flu- oroscopy images were acquired with an isotropic pixel spacing of 0.39mm, at 58kVp and 23mA with a 4ms exposure. XPlane and single plane TEE images were acquired using a Philips iE33 US machine with an X7-2t probe (operating frequency of 7 to 2 MHz) at a frame rate of 36 or 52Hz and an imaging depth of 7-10cm.
XPlane Images
Five Xplane images showing both short-axis and long-axis views were acquired in different positions covering a spectrum of images likely to be acquired in a clinical scenario. These images were registered to the CT surface mesh. Two images that did not contain a clearly visible target were not used.
Reconstructed points from multiple single plane images
Eight single plane US images were acquired simultaneously with a fluoroscopy image (to provide spatial tracking information). Two images that did not contain a clearly visible target were not used. The tracking information was used to reconstruct man- ually identified points taken from multiple US images into a common coordinate system. These points were then registered to the CT surface mesh. Images that did not contain a clearly visible target were not used. Targets were identified manually on both CT and US.
Accuracy Assessment
Accuracy of the CT-TEE and CT-fluoroscopy registrations were assessed using: 1. A mean point-to-surface distance measured between the TEE contours and CT
surface after ICP registration (CT-TEE registration only);
2. A target registration error (TRE) using five pins inserted into the aortic root as targets (CT-TEE and CT-fluoroscopy registration).
5.3.2
Results
5.3.2.1 Target localization
Targets were manually identified on all images. The standard deviation of the measured locations (determined from 10 repeated measurements for each target) were 0.41 mm, 0.40 mm and 0.23 mm for the CT, TEE and fluoroscopy images respectively. This error contributes partially to the measured target registration error.
5.3.2.2 CT-TEE Registration XPlane Images
The 165 US points identified on two planes were registered to the CT derived surface mesh, resulting in XPlane US to CT registration as shown in Figure 5.9a and b, which demonstrates good alignment of the US identified points with the segmented CT surface as shown in Figure 5.9a) and Figure 5.9b). Figure 5.9c) shows a slice through the CT mesh overlaid on its corresponding US image to visually highlight the registration error. The mean point-to-surface error and target registration error were 1.48 and 5.00mm respectively. The registrations took an average of 0.44 seconds to run on a 2.66GHz Intel Core 2 Duo processor with 4GB of memory.
Fig. 5.9: (a) Registered CT surface mesh shown overlaid onto the reconstructed US image and manually identified US points (blue). (b) US Points (blue) registered to the CT surface mesh shown with CT (green) and US (red) identified targets. (c) US image shown with corresponding CT contour (red) after registration.
Reconstructed points from multiple single plane images
The 338 US points identified on 6 planes were registered to the CT model. Results of the reconstructed US to CT registration are shown in Figure 5.10 and Table 5.2. The target locations measured using tracked TEE have a standard deviation of 0.89mm, reflecting the error introduced in the manual target selection process combined with TEE probe tracking error (Figure 5.10). There is good alignment of targets identified in CT and US (Figure 5.10b), with an RMS TRE of 2.64mm. The registrations took 0.46 seconds to run on a 2.66GHz Intel Core 2 Duo processor with 4GB of memory.
Table 5.2: Reconstructed US to CT surface registration error Point to Surface Distance (mm) TRE (mm)
N = 338 N = 6
Mean 2.12 2.48
Deviation 0.78 0.99
Fig. 5.10: (a) Pose at initialization (b) Registration output showing registered US points (blue) and CT mesh. Targets identified from CT (green), and US (red) are shown. The slight mis-alignment of the same target identified on multiple TEE images reflects manual segmentation and TEE tracking error.
5.3.2.3 CT-Fluoroscopy Registration
Targets identified on TEE were re-projected onto the fluoroscopy image and com- pared to the manual gold standard targets identified from fluoroscopy. Magnification correction was applied to the distance measurements to correct for perspective, using the height of the TEE probe (as measured using fiducial-based TEE pose estimation) as an estimate of the target distance from fluoroscopy detector. The targets demon- strated good correspondence as shown in Figure 5.11. The RMS target registration error was 5.50 mm.
Fig. 5.11: (a) CT-Fluoroscopy registration results. Red - Pin targets identified manually in the fluoroscopy image. Blue - Pin targets identified in the CT image, reprojected onto the fluoroscopy image. (b) Overlay of CT surface (aortic root and left ventricular outflow tract) onto the fluoroscopy image.
5.3.3
Discussion
The results of this study represent the accuracy of a CT-TEE registration of an aortic root without shadow artifacts (healthy porcine subjects do not have calcified aortic roots) or motion error. Results demonstrated clinically acceptable TRE for both TEE-CT and CT-fluoroscopy registration.
Points reconstructed from multiple imaging planes demonstrate an accuracy 2X greater than XPlane imaging, which is consistent with results from Section 5.2. While additional planes help improve accuracy, the use of continuous TEE probe tracking complicates the procedure. Furthermore, time is required for multiple TEE images to
be acquired, which does not allow for real-time registration and visualization. Results suggest that in many cases, XPlane imaging alone may demonstrate sufficient clinical accuracy.
XPlane TEE images demonstrate a TRE approximately 2X greater than that measured using simulated XPlane CT contours in Section 5.2, indicating that both lack of an adequate number of contours and inaccuracies introduced in the TEE imaging process are equally responsible for resultant error.
Significant error was attributed with target localization in both CT (limited by CT resolution) and TEE images (limited by TEE resolution and the presence of reverberation artifacts). The registration error measured in this experiment was likely over-estimated due to the difficulty in localizing pin targets, and better results are expected with anatomical targets, as used in Section 5.4.