The Study On Molecular Marker Detection Of Gliomas Based On MRI Analysis

Glioma is the brain tumor occurred in the neuroderm and originated from the supportive cells of the brain, called glial cells. It is one of the most commonly diagnosed primary brain tumors. At present, the treatment of the gliomas mainly depends on neurosurgery, which is supplemented by radiotherapy and chemotherapy. However, different kinds of gliomas are usually different in the clinical manifestations, the treatment plans, and prognosis. As the development of molecular marker in gliomas, neurosurgeons could develop a treatment plan according to the expression status of the molecular markers, in order to achieve a better balance between longer survival and better quality of life for the patients. Since there are still some disadvantages of the molecular marker detection method in clinical, this thesis focused on the research of the novel molecular marker detection method based on magnetic resonance image analysis.First, a novel detection method of MIB-1labeling index range based on magnetic resonance image analysis was proposed. As intensity non-uniformity degraded the quality of the images, a non-parametric method for automatic correction of intensity non-uniformity in MR images was used. In order to extract the texture features of the same area in gliomas from different sequences, the affine transformation based on a multi-resolution approach was used to map all images to one sequence. To extract the actively growing region of gliomas, a software tool called ROI_Drawing, was developed to assist neurosurgeons and radiologists in drawing the regions of interest (ROI). Then the texture features of each ROI was extracted by the gray level co-occurrence matrix, gray level gradient co-occurrence matrix, run length matrix, gradient matrix, and Minkowski functional. Finally, the support vector machine was trained as the classifier with the strategy of leave-one-out cross-validation.Based on the above strategy, a novel MRI image analysis method was proposed for the MGMT expression status detection. To avoid the subjective error in the manual drawing of ROIs, this thesis proposed a new method based on MRS measurement of Cho/NAA ratio to determine the actively growing region of gliomas. As there are a larger number of samples in the detection of MGMT expression status, feature optimization method was used to select the optimum feature subset, which was applied to evaluate the classification performance with leave-one-out cross-validation.The research results showed that the proposed method detected the range of the MIB-1labeling index efficiently, accurately and noninvasively. The features based on gray level co-occurrence matrix, run length matrix and gray level gradient co-occurrence matrix played an important role in the detection. The detection method proposed by this thesis predicted the MGMT expression status quantitatively and noninvasively. Compared with the existing two-dimensional discrete orthogonal S-transform (DOST), the proposed method was more complete and accurate. Meanwhile, a new method to extract the regions of interest (ROI) based on magnetic resonance spectrum was proposed, which made the ROI extraction more objective and easier to operate. The introduction of feature optimization improved the performance of MGMT detection not only for the original feature set, but also for the DOST method.