The Study On Brain Tumor Image Segmentation And Three-dimensional Reconstruction

The brain tumor is one of the common nervous system diseases, and whether it is benign or malignant, the tumor always presses brain tissues, harmful to the patient's physical health or even leading to death. At present, the treatment is usually based on brain CT or MRI images to get the tumor characteristics and determine whether needs operations. But the brain medical image segmentation is difficult due to the fuzzy and adhesion features of it, hardly gotting accurate results. Therefore, the research of brain tumor image segmentation method is of great significance to improve the accuracy of medical diagnosis, reduce the workload of doctors, but the three-dimensional tumor still relies on the doctor's imagination. The threedimensional reconstruction can render the shape and size of tumors, and provide more evidence for doctors. So, the research on the accurate segmentation of the brain tumor image and the three-dimensional reconstruction method has important application and practical significance to the computer assisted therapy.The study of this thesis is to segment the brain tumor image based on the C-V model optimized by watershed transformation, then using surface rendering of the marching cubes(marching cubes, MC) algorithm on the sequence of tumor image segmentation and showing the 3D reconstruction result of the tumor. The main work of this thesis is shown as follows:(1) In order to reduce the adverse effect to segmentation which is caused by edge fuzzy of the brain tumor CT images, this thesis adopts the non-linear combinations of morphological filtering based on opening and closing operations to filter the noise while maintaining the edge of the tumor characteristics as much as possible, to speed up the segmentation and reduce the segmentation errors.(2) When the brain tumor adheres to surrounding tissues, it is difficult to distinguish them by common segmentation methods alone. This thesis proposes the C-V model optimized by marker-controlled watershed to segment brain tumor images: the marker-controlled watershed segmentation result as the initial contour of the improved C-V model for curve evolution. This method can not only solve the sensibility of the initial contour problems in C-V model, receiving the segmentation of tumor from tissue adhesion, but also avoid the time-consuming re-initialization by introducing a penalty term in the energy function, reducing the large amount of calculation.(3) Aiming to display more details of tumor information, this thesis uses surface rendering to realize the three-dimensional reconstruction, with the sequence of tumor image segmentation results showing the three-dimensional features of the brain tumor, providing more evidences for clinical diagnosis.Through simulation experiments on the MATLAB2014 a platform for clinical brain tumor images, the method proposed in this thesis shows good segmentation results for common brain tumor CT/MRI images, especially for tumors with adhesion to tissues. At the same time, this thesis achieves the 3D reconstruction of the brain tumor.