Integrated reconstruction of 2D overlapping coronary artery from x-ray angiography with 3D virtual myocardial model using non-rigid iterative closest point algorithm

C-arm angiography imaging is a vital tool for the real-time visual analysis and diagnosis of coronary arteries-related diseases that occur in the coronary vessels. It is very useful for the management of coronary artery diseases. However, its two dimensional (2D) image output provides cardiologists with less accurate information. On the other hand, a three-dimensional (3D) image of these arteries will give a more precise pictorial view and facilitate a better understanding of its associated diseases. This motivates the 3D reconstruction of real coronary arteries superimposed on a 3D virtual heart model using a non-rigid iterative closest point algorithm, which is the focus of this thesis. The 3D registration involves three stages: extraction of coronary arteries centerlines after segmentation, 3D reconstruction of the coronary arteries, and appropriate computation of 3D registration between the 3D reconstructed coronary arteries and a 3D virtual heart model. In this regard, three technical problems that are related to each stage were solved. These problems include unifying the intersection points and correcting centerlines extracted from the angiographic image segments, finding the real correspondences for the reconstruction method, and 3D registration of different shapes. This thesis presents three vital contributions to these major aspects of the 3D registration of coronary arteries with a 3D virtual heart model. A fully automated method for improving the accuracy of the following processes: correcting the centreline extracted from the coronary artery tree segment, separating overlapping vessels based on the B-spline method, and producing a multilayer image. The average mean square error (MSE), accuracy, sensitivity, precision, and specificity values of the proposed method using clinical datasets are 2.22%, 97.7%, 83.3%, 83.3%, and 98.8%, respectively. Using the synthetic dataset, the obtained values are 2.54%, 98.2%, 85.6%, 84.8%, and 98.9%, respectively. These results show that the efficiency of the proposed method. Specifically, the proposed method has low MSE. It also enhances the performance of coronary artery segmentation methods in angiography by solving the problem of overlapping vessels. In other words, it aids the segmentation of coronary arteries in X-ray angiographic images without avoiding angiograms with overlapping vessels. In the reconstructed coronary artery tree. Its peculiar advantage lies in the 3D reconstruction of the coronary artery tree from X-ray angiographic images without the need for corresponding points between vessels in each view. The results of the experiments performed using a clinical dataset prove that the proposed method decreases the mean errors of the right coronary artery to 0.1465 and the left coronary artery to 0.2453. Moreover, since there is no need to select correspondence for each point between the angiograms, the proposed method solves a unique problem within this subject area. The proposed algorithm to register different shapes, i.e., the 3D coronary arterial vessels and the 3D virtual heart model. The proposed algorithm was affirmed by computer graphics experts, cardiologists, radiologists, and patients in a questionnaire survey. These research contributions are applicable in practice and can be conveniently deployed using personal computers and standard medical acquisition techniques without any change in the medical acquisition standards. In other words, there is no needed to calibrate acquisition devices. This makes the procedure easy to deploy on most X-ray angiography devices.

Text FSKTM 2020 9 ir.pdf Download (1MB)

Actions (login required)

View Item