Automated segmentation and uncertainty quantification for image-based cardiovascular modeling with convolutional neural networks
Abstract/Contents
- Abstract
- In this thesis we accelerate and extend a path-planning based patient-specific modeling method commonly used for anatomic model creation for cardiovascular fluid dynamics simulations. We further address a longstanding open question of how realistic patient-specific model geometry variability influences simulation output uncertainty. Model building is accelerated by using recently developed deep learning methods and convolutional neural networks to automatically generate vessel surfaces from image data. We enable the quantification of simulation output uncertainty due to geometry variation by modeling the probability distribution of vessel surfaces using convolutional Bayesian dropout networks. In the first part of this thesis we use fully-convolutional neural networks (FCNN) to generate 2D cardiovascular segmentations. Given vessel pathlines, the neural network generates 2D vessel enhancement images along the pathlines. Thereafter, vessel segmentations are extracted using the marching-squares algorithm, which are then used to construct 3D cardiovascular models. The neural network is trained using a novel loss function, tailored for partially labeled segmentation data. An automated quality control method is also developed, allowing promising segmentations to be selected. Thereafter we develop machine learning models to directly predict vessel lumen surface points using a regression formulation. In contrast to the previous method, which identifies the vessel lumen through binary pixel classification, formulating vessel lumen detection as a regression task allows predictions to be made with human expert level accuracy. The regression formulation enables machine learning models to be trained end-to-end for vessel lumen detection without post-processing steps that reduce accuracy. In the third part of this thesis we propose a novel approach to sample from a distribution of patient-specific models for a given image volume. The method uses the previously developed vessel lumen regression networks, combined with dropout layers, to enable Bayesian sampling of vessel geometries. The networks are then applied in the path-planning patient-specific modeling pipeline to generate families of cardiovascular models. A key innovation is the ability to learn geometric uncertainty directly from training data based on medical images. We then quantify geometric uncertainty for clinically relevant anatomies, and provide detailed analysis of its effects on cardiovascular patient-specific fluid dynamics simulation results. The above methods allow for efficient automation of patient-specific model construction from medical images which greatly accelerate the model construction process and reduce laborious user input. These methods are combined with uncertainty quantification methods that enable assessment of how image-based uncertainty propagates to simulation outputs
Description
| Type of resource | text |
|---|---|
| Form | electronic resource; remote; computer; online resource |
| Extent | 1 online resource |
| Place | California |
| Place | [Stanford, California] |
| Publisher | [Stanford University] |
| Copyright date | 2020; ©2020 |
| Publication date | 2020; 2020 |
| Issuance | monographic |
| Language | English |
Creators/Contributors
| Author | Maher, Gabriel Dominic |
|---|---|
| Degree supervisor | Marsden, Alison |
| Thesis advisor | Marsden, Alison |
| Thesis advisor | Gerritsen, Margot (Margot G.) |
| Thesis advisor | Rubin, Daniel (Daniel L.) |
| Degree committee member | Gerritsen, Margot (Margot G.) |
| Degree committee member | Rubin, Daniel (Daniel L.) |
| Associated with | Stanford University, Institute for Computational and Mathematical Engineering. |
Subjects
| Genre | Theses |
|---|---|
| Genre | Text |
Bibliographic information
| Statement of responsibility | Gabriel Maher |
|---|---|
| Note | Submitted to the Institute for Computational and Mathematical Engineering |
| Thesis | Thesis Ph.D. Stanford University 2020 |
| Location | electronic resource |
Access conditions
- Copyright
- © 2020 by Gabriel Dominic Maher
- License
- This work is licensed under a Creative Commons Attribution Non Commercial 3.0 Unported license (CC BY-NC).
Also listed in
Loading usage metrics...