Usage
  • 1 view
  • No download information available

A Diffeomorphic 3D-to-3D Registration Algorithm for the Segmentation of the Left Ventricle in Ultrasound Sequences

  • Author / Creator
    Krishnaswamy, Deepa
  • Heart disease is the second leading cause of death in Canada, where it affects the lives of over two million people. One modality used to detect and diagnose heart disease and other abnormalities is echocardiography or ultrasound imaging of the heart. Ultrasound imaging, compared to other modalities has several advantages; it is non-ionizing, portable, and cost-effective, and provides good spatial and temporal resolution. It is crucial that the left ventricle must be analyzed in the case of cardiac diseases. Metrics derived from analysis of the left ventricle provide an indication to the clinician about the performance of the heart. However, the current clinical software and methods available in the literature to analyze the left ventricle suffer from several potential drawbacks. Geometrical assumptions may be made about the chamber, or a large amount of manual interaction is required. In the case of supervised deep learning neural networks, a training dataset may be required, which may be difficult to obtain. Therefore the goal of this thesis was to focus on the development of semi-automated methods to delineate the endocardium of the left ventricle based on registration. The methods developed do not require the use of training data, geometrical assumptions, or prior knowledge about the image characteristics. The thesis focuses mainly on the application to ultrasound sequences, with additional testing on MR sequences. In particular, a semi-automated method has been developed with the use of a diffeomorphic registration algorithm to delineate the endocardial borders at end-diastole and end-systole. This method was expanded to provide a segmentation over the full temporal sequence of ultrasound images. Lastly, a 3D-to-3D diffeomorphic registration method was developed for segmentation, where the algorithm was able to capture the full dynamics of the motion of the left ventricle over the cardiac cycle. We have compared the proposed methods to other common registration packages in terms of standard distance and clinical metrics. The results demonstrate the benefit of using a diffeomorphic registration method for the segmentation of the left ventricle.

  • Subjects / Keywords
  • Graduation date
    Fall 2021
  • Type of Item
    Thesis
  • Degree
    Doctor of Philosophy
  • DOI
    https://doi.org/10.7939/r3-7xbp-te27
  • License
    This thesis is made available by the University of Alberta Libraries with permission of the copyright owner solely for non-commercial purposes. This thesis, or any portion thereof, may not otherwise be copied or reproduced without the written consent of the copyright owner, except to the extent permitted by Canadian copyright law.