Electronic Thesis and Dissertation Repository

Thesis Format

Integrated Article


Doctor of Philosophy


Biomedical Engineering


Peters, Terry M.

2nd Supervisor

Chen, Elvis C.S.



Mitral valve disease is a common pathologic problem occurring increasingly in an aging population, and many patients suffering from mitral valve disease require surgical intervention. Planning an interventional approach from diagnostic imaging alone remains a significant clinical challenge. Transesophageal echocardiography (TEE) is the primary imaging modality used diagnostically, it has limitations in image quality and field-of-view. Recently, developments have been made towards modelling patient-specific deformable mitral valves from TEE imaging, however, a major barrier to producing accurate valve models is the need to derive the leaflet geometry through segmentation of diagnostic TEE imaging. This work explores the development of volume compounding and automated image analysis to more accurately and quickly capture the relevant valve geometry needed to produce patient-specific mitral valve models.

Volume compounding enables multiple ultrasound acquisitions from different orientations and locations to be aligned and blended to form a single volume with improved resolution and field-of-view. A series of overlapping transgastric views are acquired that are then registered together with the standard en-face image and are combined using a blending function. The resulting compounded ultrasound volumes allow the visualization of a wider range of anatomical features within the left heart, enhancing the capabilities of a standard TEE probe.

In this thesis, I first describe a semi-automatic segmentation algorithm based on active contours designed to produce segmentations from end-diastole suitable for deriving 3D printable molds. Subsequently I describe the development of DeepMitral, a fully automatic segmentation pipeline which leverages deep learning to produce very accurate segmentations with a runtime of less than ten seconds. DeepMitral is the first reported method using convolutional neural networks (CNNs) on 3D TEE for mitral valve segmentations. The results demonstrate very accurate leaflet segmentations, and a reduction in the time and complexity to produce a patient-specific mitral valve replica. Finally, a real-time annulus tracking system using CNNs to predict the annulus coordinates in the spatial frequency domain was developed. This method facilitates the use of mitral annulus tracking in real-time guidance systems, and further simplifies mitral valve modelling through the automatic detection of the annulus, which is a key structure for valve quantification, and reproducing accurate leaflet dynamics.

Summary for Lay Audience

Three-dimensional ultrasound is widely used for obtaining images of the heart for both preoperative-diagnostic and intraoperative-guidance purposes. For surgical procedures targeting the mitral valve, which controls the flow of blood from the left atrium to the left ventricle, 3D ultrasound images are acquired from a probe inserted into the esophagus which provides clear 3D images of the mitral valve and surrounding tissues. Our objective is to develop and validate systems that leverage advanced image processing approaches to improve information from diagnostic ultrasound and use this information for training and planning interventions. Ultrasound imaging of the heart can show some features in a high level of detail, like the mitral valve, but is very limited in use for important structures that lay in the left-ventricle. I propose a workflow for acquiring and compounding, or stitching, multiple separate 3D images together to reconstruct a single image showing the entire left side of the heart in a process similar to a panoramic picture, enabling clinicians to better plan for procedures. Additionally, automatic image segmentation, or labelling, will be used in a workflow for creating patient-specific silicone replicas of the mitral valve, which surgeons can use to practice procedures and compare various approaches to aid in the planning process. This will have applications in both surgical planning, as well as training, by providing a platform for clinical users to practice and evaluate the successfulness of a procedure.

Creative Commons License

Creative Commons Attribution-Share Alike 4.0 License
This work is licensed under a Creative Commons Attribution-Share Alike 4.0 License.