Development of a registration framework for echocardiography and computed tomography cardiovascular image fusion / Azira Khalil

• Main Author: Azira, Khalil
• Format: Thesis
• Published: 2018
• Subjects: R Medicine (General); T Technology (General)
• Online Access: http://studentsrepo.um.edu.my/8449/2/All.pdf; http://studentsrepo.um.edu.my/8449/6/azira.pdf; http://studentsrepo.um.edu.my/8449/

Echocardiography and Cardiac CT images are important imaging modalities in diagnosing the cardiac diseases. The image registration and fusion could offer the solution to the physician where the integrated image from these two modalities provide complementary information. This study proposed a registration framework to register 2D echocardiography images with 3D cardiac CT images. The registration facilitates the fusion of CT and echocardiography to aid the diagnosis of aortic and mitral valves diseases. It provides surgical guidance during transcatheter valve replacement (TVR) and implantation (TVI). The developed image registration framework consists of three major steps: temporal registration, noise reduction and spatial registration. Temporal registration allows time stamping of echocardiography time series data to identify frames that are at a similar cardiac phase as the cardiac CT volume data. Next, in the preprocessing phase, the noise of echocardiography image is reduced using the speckle reducing anisotropic diffusion (SRAD) technique. For the spatial registration, an intensity-based normalized mutual information (NMI) method is applied with pattern search optimization algorithm to produce an interpolated cardiac CT image. This registration framework however does not utilize the optical tracking information. The proposed registration method has been applied on the short axis "Mercedes Benz" sign, long parasternal axis and four chamber views of cardiac images from ten patients. The accuracy of the fully automated registration method is 0.81 ± 0.08 and 1.30 ± 0.13 mm in terms of the Dice Similarity Coefficient (DSC) and the Hausdorff distance (HD) for aortic valve respectively. Besides that, for the long body of aorta, the DSC and HD are 0.79 ± 0.02 and 1.19 ± 0.11 mm respectively. DSC and HD for the left atrium assessments are 0.87 ± 0.04 and 1.23 ± 0.32 mm respectively, whereas for left ventricle are 0.82 ± 0.07 and 1.14 ± 0.18 mm respectively. These accuracies are comparable to gold standard manual registration by expert. There was no significant difference in aortic and mitral annulus diameter measurements between the automatically and manually registered CT images. The transformation parameters showed small deviation (RMSE ≤ 2.12 mm deviation in translation and <3º for rotation) between manual and automatic registration. Without the use of optical tracking, the registration method has shown its applicability for effective fusion of echocardiography with preoperative CT volume data to potentially facilitate catheter-based surgery. Thus, the proposed method enables to aid the physician in diagnosing the aortic and mitral valves disease as well as providing surgical guidance during the treatment procedure of TVR or TVI.