An Improved Edge Detection for Left Ventricle Epicardial and Endocardial on Short Axis Magnetic Resonance Images
| dc.contributor.author | Eileen Yap Pin Pin | |
| dc.date.accessioned | 2026-10-15T03:36:40Z | |
| dc.date.issued | 2018 | |
| dc.description | Cardiac Magnetic Resonance Imaging (MRI) is a useful technology for medical doctors to diagnose heart diseases, however the automation of the cardiac segmentation application is still challenging due to the complex structures inside and around the heart. The previous research works have focused on locating the heart using training data sets, however they have yet to establish an effective thresholding method. Therefore, this research proposed a method with the aim of creating an effective and robust threshold to improve image segmentation in pre-processing stage, calculate the thresholds based on the user defined region-of-interest (ROI), and have a segmentation initialization mask closest to the defined ROI. A data set containing a set of 10 images were obtained from the Sarawak Heart Centre for testing purposes and the Sunnybrook Cardiac MR image data sets were used for validation and benchmarking stage. A localized region-based active contour algorithm proposed by Lankton (2009) was adapted for image segmentation. An evaluation was done through subjective assessment by a cardiologist and automated evaluation measurements by using the 2D image correlation, sensitivity, specificity and dice metric measures. The prototype testing has shown to be easy to use, after minor adjustments to the input image size. Optimal thresholding for test data set was 150 pixel intensity for pre-processing and mean ± one standard deviation for thresholding. Validation noted parameters of 0.9 (or 90%) and above and benchmark with Dice metric (epicardial edge: 0.93 ± 0.05; and endocardial edge: 0.92 ± 0.04) were comparable with other studies such as Huang et al. (2009), Lu et al. (2009) and Hu et al. (2014). The proposed methods in this study were successful in removing unwanted noise structures to produce accurate edges. The limitations of the current methods include the need for detecting round shapes for left ventricle in the images. Further research v is needed to refine the methods to determine the optimal thresholding level and the adaptation for more complex shapes. | |
| dc.identifier.uri | https://scholarhub.unimas.my/handle/123456789/3100 | |
| dc.language.iso | English | |
| dc.publisher | UNIVERSITI MALAYSIA SARAWAK | |
| dc.relation.ispartofseries | Faculty of Computer Science and Information Technology | |
| dc.subject | MRI, image processing, edge detection, left ventricle, heart | |
| dc.title | An Improved Edge Detection for Left Ventricle Epicardial and Endocardial on Short Axis Magnetic Resonance Images | |
| dc.type | Masters |