The registration of the preoperative 3D coronary model (3DCM) to the intraoperative patient environment will allow the accurate positioning of the virtual cardiac model to the beating heart in the patient’s thorax.
Generation of simulated angiogram with background by the addition of the DRR image to a real patient angiogram background retrieved from the patient 224 study. (a) Simulated angiogram with no background generated by the DRR technique. 60° LAO view of the LCA. (b) Real patient angiogram background or precontrast angiogram retrieved from patient 224 study. (c) Simulated angiogram with background resulting from the fusion of image (a) and (b).
Pixel value similarity comparison between a simulated and a real angiogram. (a) Simulated angiogram with background of the LCA at 60° LAO view from dataset 6 created by the addition of a 4DBM DRR and patient 224 background (Fig. 2). (b) Clinical angiogram of patient 224 with contrast agent. (c) Resulting SABM of (a) after multiscale analysis. (d) Resulting binary mask image of (b) after multiscale analysis.
2D-3D registration of a preoperative 3DCM to two intraoperative angiograms that have been automatically segmented into binary masks.
Comparison of a MBP as generated at each registration iteration and of a SABM. (a) MBP of the preoperative mesh 3DCM as generated at each registration iteration. LCA 60° LAO. (b) Simulated angiogram with background of the LCA at 60° LAO view from dataset 1 processed with the multiscale segmentation yielding the binary mask image or SABM. Note that the extra “branch” at the top of (b) (arrow) is a catheter that is part of the background, rather than part of the vascular system.
3DRMS error calculated between homologous points of the registered coronary tree and the gold standard. Accuracy results of the dataset without anatomical background and the eight “patient” datasets for: 1P and 2P ideal cases, 2P with time offset, and 2P with time offset and reconstruction noise.
Plot of the cost function for each of the , , and projection axes for LCA dataset 1, single plane 60° LAO view. For the in-plane axis ( and ), the cost function peak is at and for the out-of-plane axis , it lies at .
3D plot of dataset 8, single plane cost function for the 30° LAO view of the LCA. Cost function values for in-plane, and , at the out-of-plane offset of .
Nine datasets used to validate the registration. Dataset 0 is composed of DRR images without the addition of anatomical background. Datasets 1–8 are produced from the combination of DRR images and real patient angiogram backgrounds taken at matching acquisition angles (Fig. 2). Column 2 identifies the patient number from which the real patient angiogram backgrounds were taken, column 3 lists the arterial branch, and columns 4 and 5 present the RAO and the LAO angulation of the views used for the registration. The LAO view is used for the single-plane registration.
Results of the 2D-3D registration experiments. Column 3 gives the total number of registrations performed using Table I datasets and the ten starting positions. Column 4 lists the average number of restarts necessary for registration success. Column 5 gives the 3DRMS error of the successful registrations. Columns 6 and 7 present the time taken to perform the registration and the success rate, respectively.
Article metrics loading...
Full text loading...