Dissertation
A procedure for multimodal brain tumor segmentation
University of Iowa
Doctor of Philosophy (PhD), University of Iowa
Spring 2022
DOI: 10.25820/etd.006540
Abstract
Interpretable, accurate, and reproducible brain tumor segmentation approaches are crucial to diagnosis, treatment planning, and follow-ups of brain tumors. A three-stage procedure for multimodal brain tumor segmentation is proposed in the thesis. The procedure segments a glioma patient’s brain into healthy tissue types and tumor tissue types by assigning labels to each volume element. The contextual constraints on the tissues are modeled by a Potts model and the interactions among observed grayscale values are modeled by a Gaussian Markov random field (GMRF). Instead of treating all types of tissue equally as in conventional hierarchical MRF models, independent and correlated bright signals are defined. Disjoint regions identified as correlated bright signals are allowed to have independent GMRF parameters. An iterated conditional modes algorithm is used to find the local maxima of the posterior distributions. The different modalities of magnetic resonance images are preprocessed
and segmented in a sequence where the T1ce image is preprocessed and segmented first, the FLAIR image is preprocessed and segmented second, and the T2 image is preprocessed and segmented last. A postprocessing approach is also proposed to yield final segmentation results. The performance of the proposed procedure has been evaluated on the data provided by the Multimodal Brain Tumor Segmentation Challenge (BraTS 2019). The experimental results show that the proposed procedure can yield accurate brain tumor segmentation.
Details
- Title: Subtitle
- A procedure for multimodal brain tumor segmentation
- Creators
- Hongda Zhang
- Contributors
- Luke Tierney (Advisor)Brian J. Smith (Committee Member)Mary Kathryn Cowles (Committee Member)Aixin Tan (Committee Member)Joyee Ghosh (Committee Member)
- Resource Type
- Dissertation
- Degree Awarded
- Doctor of Philosophy (PhD), University of Iowa
- Degree in
- Statistics
- Date degree season
- Spring 2022
- DOI
- 10.25820/etd.006540
- Publisher
- University of Iowa
- Number of pages
- xiii, 123 pages
- Copyright
- Copyright 2022 Hongda Zhang
- Language
- English
- Description illustrations
- illustrations (some color)
- Description bibliographic
- Includes bibliographical references (pages 116-119).
- Public Abstract (ETD)
Magnetic resonance imaging (MRI) is a non-invasive medical imaging technique for generating three dimensional (3D) anatomical images of soft tissues in the body. In the brain, MRI can help distinguish healthy tissues and a variety of tumor tissues. The process of splitting a brain image into healthy tissues and tumor tissues is brain tumor segmentation. Brain tumor segmentation is challenging since the shapes and appearances of tumor regions vary in a brain. Interpretable, accurate, and reproducible approaches for brain tumor segmentation are in demand because they are crucial to diagnosis, treatment planning, and follow-ups of brain tumors. A novel three-stage procedure, which consists of preprocessing, processing, and post-processing stages, is proposed in the thesis for brain tumor segmentation. Instead of being analyzed simultaneously, the MR images are preprocessed and processed in a sequence. Then the intermediate results are combined in the postprocessing stage. Experiments have been conducted on the data from the Multimodal Brain Tumor Segmentation Challenge (BraTS 2019). The results demonstrate that the proposed procedure can produce accurate brain tumor segmentations.
- Academic Unit
- Statistics and Actuarial Science
- Record Identifier
- 9984271155602771
Metrics
3 File views/ downloads
204 Record Views