| Glioma is the most common tumor in brain,which seriously threatens people's health.Clinical experiments have shown that methylation is one of the important factors to ensure the effect of chemotherapy.In order to reduce the cost of methylation detection,reduce the suffering of patients and improve the effect of chemotherapy,this paper uses a method of pattern recognition to classify methylation,and through the establishment of classification model to achieve the prediction of methylation,thus assisting clinical doctors to make the best treatment plan.In this paper,we establish a methylation classification model by mining imaging information from CT,MRI.The main tasks of this paper are:(1)Collecting image data of 216 patients from 2011 to 2017 and preprocessing the images.(2)Using medical image processing software MITK to perform semi-automated segmentation of CT and MRI images of patients with glioma,and to propose regions of interest.(3)Extracting high-dimensional image features from the regions of interest,including first-order statistics,wavelets,textures,shapes and sizes.(4)Taking the extracted image feature as an independent variable,the methylation level as a dependent variable to compose a new data sequence for subsequent data selection.(5)Using Lasso regression to select features and using the “cross-validation” to find the optimal value of parameter ?,so that the optimal feature subset is selected from the data sequence and the data dimension and redundancy are reduced.(6)Using the “leave-out” to divide feature subsets into training samples and test samples,using training samples to build a methylation classification model based on Support Vector Machines(SVM),and then using test samples to evaluate the accuracy of the model.Using the methods described in the article,4356 image features are extracted from the images of each patient,and a classification model for methylation of gliomas is established.The accuracy rate of the model is 72.727%,which has a certain clinical aids. |
PDF Full Text Request