Loading...

Scar Segmentation in CMR Images without Using Contrast Agent

Badali Golezani, Elaheh | 2022

175 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 55055 (19)
  4. University: Sharif University of Technology
  5. Department: Computer Engineering
  6. Advisor(s): Rohban, Mohammad Hossein; Houshmand, Golnaz
  7. Abstract:
  8. Correct diagnosis of myocardial scar has always been a major challenge due to the low resolution of cardiac magnetic resonance imaging. The use of a gadolinium-based contrast agent that reveals a scar is the solution proposed in medical science. However, there are limitations to the use of contrast agents in some patients. In recent years, studies based on deep learning techniques have been presented, trying to identify myocardial infarction with the help of images without using contrast agent. This type of diagnosis can be done with the help of different movements of healthy and damaged tissue. Due to the lack of datasets suitable for this application, in this study, real dataset were received and appropriate dataset were created after solving the existing challenges. Then, different networks were examined for the scar segmentation. The best model that produced appropriate results on this real dataset is the one based on the 3D U-Net model, to which some sections have been added to improve the results. One of the innovations made is the use of the VoxelMorph network to extract kinetic information between different slice of images and to add this information to the information extracted by U-Net network. Also, in another structure, the ResU-Net model was used as the base network instead of using the simple U-Net model, and VoxelMorph information was added to it, which showed the best result among all the tested models. In this study, Dice score has been used as an evaluation metric. Finally, based on 4 of the best models, a ensemble decision was made, which greatly improved the results. The Dice score of this model is 59.38%.
  9. Keywords:
  10. Deep Learning ; Cardiac Scar Segmentation ; U-Net Model ; Gadolinium Contrast Agent ; Cardiac Cine Magnetic Resonance Imagin (MRI)

 Digital Object List

 Bookmark

No TOC