Detection of myocardial infarction in 2D echocardiograms and cardiac magnetic resonance images using deep learning | |
| Author | Ballais, Lalaine Jean Aragon |
| Call Number | AIT Thesis no.TC-22-02 |
| Subject(s) | Deep learning Echocardiography, Two-Dimensional Coronary heart disease--Diagnosis imaging |
| Note | A thesis submitted in partial fulfillment of the requirements for the degree of Master of Engineering in Telecommunications |
| Publisher | Asian Institute of Technology |
| Abstract | Ischemic heart disease (IHD) is among the world's principal cause of death and disabil- ity. In the Philippines alone, it is claimed that approximately 99.7 thousand Filipinos died in the year 2020 due to IHD and outranked the number of deaths due to any other disease, including Co ViD-19. IHD, when prolonged could result to the death of heart muscle cells, a condition known as myocardial infarction (MI). Many people expe- rience IHD without symptoms leaving it untreated and suddenly suffering from my- ocardial infarction. According to the global epidemiology for IHD, one of the clinical manifestations of IHD is myocardial infarction. Therefore, finding a way to diagnose myocardial infarction early is of great importance and is the subject of research for many scholars. Myocardial infarction can be diagnosed through imaging techniques such as the echocardiograms and cardiac magnetic resonance images (MRI). Analyz- ing echocardiograms and MRIs are manually done by radiologists and therefore time- consuming and an experience-dependent task. In both imaging modalities, quantifica- tion techniques using statistical signal processing and deep learning algorithms can be used in the detection of myocardial infarction. Quantification techniques, however, re- main a challenge due to the variability in image quality as well as variability in cardiac structures across different subjects. In this study, myocardial infarction is detected by quantifying left ventricle wall motion and myocardial thickening in echocardiograms and MRIs by applying statistical signal processing in a combination of attention U-Net and modified siamese neural network based deep learning architecture. In both imag- ing modalities, the measurement methods were compared. Then, the technique was compared to existing ones. Results show better performance on segmentation and MI detection on echocardiogram than on MRI. |
| Year | 2022 |
| Type | Thesis |
| School | School of Engineering and Technology (SET) |
| Department | Department of Information and Communications Technologies (DICT) |
| Academic Program/FoS | Telecommunications (TC) |
| Chairperson(s) | Attaphongse Taparugssanagorn |
| Examination Committee(s) | Teerapat Sanguankotchakorn;Chaklam Silpasuwanchai |
| Scholarship Donor(s) | Asian Development Bank - Japan Scholarship Program (ADB-JSP) |
| Degree | Thesis (M. Eng.) - Asian Institute of Technology, 2022 |