ISSN 2756-3324
African Journal of Medicine and Surgery ISSN 2756-3324 Vol. 11 (7), pp. 001-023, July, 2024. Available online at www.internationalscholarsjournals.org © International Scholars Journals
Full Length Research Paper
Automated Segmentation and Extraction of Coronary Artery Tree Structures from Angiograms Using Gaussian Filtering and Thresholding
Hasan H. Khaleel1*, Rahmita O. K. Rahmat1, D. M. Zamrin2, Ramlan Mahmod1 and Norwati Mustapha1
1Department of Multimedia, Faculty of Computer Science and Information Technology, UPM, 43400, Serdang, Malaysia.
2Heart and Lung Centre, National University of Malaysia Medical Centre, HUKM, 56000 Cheras, Kuala Lumpur, Malaysia.
Accepted 24 April, 2024
Abstract
Dynamic variations in the curvilinearity of coronary arteries pose significant challenges for study using current angiograms. Nonetheless, both experimental and clinical evidence indicates that vessel extraction is valuable for surgical treatment and clinical research. In this paper, we present an automated algorithm designed to identify the outlines of coronary artery tree blood vessels in angiograms. This approach serves as a practical tool for physicians. The algorithm automates the segmentation of coronary arteries from cineangiograms, followed by precise extraction of vessel features. Such preprocessing, in conjunction with a matched Gaussian filter, can significantly enhance results. The segmentation algorithm comprises two key processes: (1) Gaussian filtering of blood vessels and (2) thresholding. We evaluated the algorithm using a raw dataset of 100 angiogram images, validating the results through two methods. First, hand-labeled annotations provided ground truth segmentation for 20 images, revealing that our algorithm outperformed manual detection, even in cases of poor contrast that the naked eye could not recognize. Second, we employed a questionnaire to assess the effectiveness of the illustrated output. The hand-labeling matched our results with an accuracy of 98%, while the questionnaire validation rate was 90.84%. We conclude that our enhanced algorithm is effective for extracting coronary artery tree vessels, including smaller branches. Additionally, the algorithm operates efficiently, completing vessel extraction in approximately 14 to 15 seconds per image.
Key words: Angiocardiography, coronary artery segmentation, matched filter, adaptive thresholding, vessel extraction.