Research On The Image Classification Of Brain Glioma Based On Machine Learning

Accurate classification of tumor have important influence on treatment for patient.Glioma has characteristic of heterogeneous,and there is no effective means to comprehensive and quantitative evaluate the heterogeneity for glioma.Because of the lack of valid information to provide guidance for doctor,ultimately lead to clinical treatment not good.Compared with traditional clinical diagnose for disease,radiomic is a multidisciplinary and have advantage of non-invasion、quantification,and radiomic applied in medical research.Radiomic from Magnetic Resonance Imaging(MRI),Positron Emission Tomography(PET)images and X ray Computed Tomography(x-CT)imges etc.extract large、quantitative high dimensional features,thus complete medical image data turn to computer can use.At present,few studies have been done on image classification of brain glioma.In this paper,Random forest(RF)algorithm and Support Vector Machine(SVM)algorithm are used to establish the classification model of glioma.In this paper,71 case of simple glioma Computed Tomography(CT)image and clinical information were selected from People’s Hospital of Henan Province,from image extracted features,using Minimal Redundancy Maximal Relevance(MRMR)and RF reduce dimensional of features.Two methods of machine learning SVM and RF were used to classify the gliomas.The correct classification rate of the Random Forest method on training set is 77%,and correct classification rate on test set is 69.6%,AUC is 0.7572.The correct classification rate of Support Vector Machine method on training set is 80%,and correct classification rate on test set is 73.9%,AUC is 0.8253.The experimental results show that SVM model has better classification result,compared with the traditional clinical pathological characterization of tumor,the image features extracted in this paper can describe the glioma accurately and quantitatively,some image features has reference value on clinical diagnosis for doctor.The data reduction experiment and the prediction model of glioma grading are effective,established predictive model can be used to assist doctor in the diagnosis benign or malignant of glioma.