QUANTITATIVE DIGITAL CORONARY ANGIOGRAPHY

Abstract

An accurate assessment of coronary arterial dimensions is essential in the clinical evaluation and management of patients with ischemic heart disease. Conventional methods for assessing stenosis of the arterial diameter rely on visual examination of the angiograms, which are X-ray images of contrast-filled coronary arteries. Such simple evaluations often result in large inter-observer and intra-observer variability in the estimates. The digitization of these angiograms into computer images allows computer-assisted tracking and border detection procedures to be applied for the quantitative analysis of these images. Two aspects of the quantitative analysis of digital coronary angiograms are analyzed in this project. The first one is the automated analysis of the angiograms to obtain the arterial width profile and the second is the compression of these digital images for efficient storage and transmission. The methods for the accurate identification of the arterial centerline and its border have been accomplished, with minimal operator interaction. The modified beam algorithm is a centerline tracking algorithm that is both fast and with a success score of 90% when tested with images of varying qualities. In the case where images are of very low contrast, another centerline tracking algorithm, the graph searching method is able to found to be able to solve the problem. Next, a template matching method uses the Gaussian convolution masks to extract features of the major arteries in the angiograms is developed. It requires the least operator interaction and suggests a promising way to fully automate the analysis procedures with no operator interaction at all. Finally, a Gaussian function algorithm that makes use of the flexibility and robustness of the Modified Beam algorithm and the Template Matching methods is developed for the fast tracking of centerline of known orientation. After the detection of the arterial centreline, a two-dimensional convolution edge detection method to detect the border of the artery is developed. This method provides a much better solution in the case of noise prone and low contrast images when compared with other conventional methods. A cineangiogram analysis algorithm is then developed to analyse sequential angiograms. in which the operator need to indicate two points on the first frame of the sequence only. Reversible techniques (lossless) for both intraframe and interframe compression are explored and developed for their suitability in reducing bit rate for the storage of the digital angiograms. First, techniques that exploit the statistical redundancy of the images, i.e. the Huffman coding and the arithmetic coding, are implemented. Then, four techniques of intraframe de-correlation of images are developed. It is found that average-DPCM method, a new de-correlation method preforms better than the other three, giving an average entropy of 3.372 bits/pixel. Frame difference, interframe DPCM and motion compensation techniques are implemented for the interframe techniques. It is found that only a hybrid method using motion compensation and average-DPCM gives better result than the intraframe average-DPCM method. The entropy is reduced to 3.277 bits/pixel.