Loading...
Glioma Tumor Segmentation in Brain MRI Using Atlas-based Learning and Graph Structures
Barzegar, Zeynab | 2020
584
Viewed
- Type of Document: M.Sc. Thesis
- Language: Farsi
- Document No: 53601 (19)
- University: Sharif University of Technology
- Department: Computer Engineering
- Advisor(s): Jamzad, Mansour; Beigy, Hamid
- Abstract:
- Brain cancer is a lump or tumor in the brain caused by abnormal growth of cells. Glioma is a common type of tumor that develops in the brain. In order to plan precise treatment or accurate tumor removal surgery, brain tumor segmentation is critical for detecting all parts of tumor and its surrounding tissues. To visualize the brain anatomy and detect its abnormalities, we use Magnetic Resonance Imaging (MRI) as an input. Due to many differences in the shape and appearance, accurate segmentation of glioma for identifying all parts of the tumor and its surrounding tissues in cancer detection is a challenging task. Moreover, due to the intensity inhomogeneity existing in brain MRI and gray level similarity between normal tissues and tumors, achieving an acceptable accuracy rate needs more biological information.The aim of this research is to design and implement an efficient and automated segmentation model for brain tumor segmentation and cancerous tissues identification using machine learning and image processing algorithms.Since the cancerous tissues (in case of existence) are less than the normal tissues in the medical images, we proposed an algorithm that selects a subset of image voxels as the labeling candidates using symmetry of the left and right hemispheres of the brain. Therefore, the search space is extremely reduced. This algorithm is used as a pre-process step in our main models.We addressed the brain tumor segmentation problem from two general approaches. The first approach is based on generative models. The focus of these models is on extracting the features which include 3D information of input images. We presented two effective generative models in the first part of this study. The experiment results show that these two proposed generative models are accurate enough. Besides, using prior knowledge about glioma tumor can improve the accuracy.The second approach is based on using prior knowledge about glioma in the form of a probabilistic model. In recent researches, combinations of atlas-based segmentation and machine learning methods have presented superior performance over other automatic brain image segmentation algorithms. In this study, we extracted prior knowledge from atlas images. In addition, in order to represent the neighborhood effect in the efficient way, we proposed a probabilistic structure in the form of a graph. Among the segmentation methods, graph-based approaches are powerful tools due to their ability in reflecting global image properties. In addition, they reduce computational complexity of segmentation problem. Results show that using appropriate prior knowledge can help to obtain better estimates of labels, and thereby better segmentation results. Performance of the proposed method is evaluated on BRATS datasets provided by the Multimodal Brain Image Segmentation Challenge. All BRATS datasets are publicly available and they include four MRI sequences (T1, T1C, T2, and T2Flair). The experimental results show that the proposed models significantly improved the segmentation result and achieved higher accuracy
- Keywords:
- Probabilistic Graphical Models ; Cancer Tissue ; Magnetic Resonance Imagin (MRI) ; Brain Tumor ; Glioma Tumor ; Multi Atlas Segmentation ; Three Dimentional Neighborhood
- محتواي کتاب
- view