Prediction Of Benign And Malignant Of Lung Nodules Based On Quantitative Radiomic Method

Cancer seriously threatening the health of mankind around the world, lung cancer is most common and deadly cancer in all cancers. The early diagnosis of lung cancer, which is crucial to the treatment, can improve the survival rate of patients. The biopsy is the main diagnostic method to assist clinical doctor discriminate the benign and malignant of lung tumor. It requires invasive surgeries to extract and analyses what are generally small portions of the entire tumor tissue. As lung tumors are spatially and temporally heterogeneous, this limits the use of invasive techniques,but the medical imaging can capture intratumoural heterogeneity. There is a lack of predicting the benign and malignant of lung nodules in a non-invasive way with a relative high accuracy.Radiomics is an emerging field that generates models which can describe the tumor and clinical manifestations and the relation between clinical phenotype and image features through extracting large amounts of advanced quantitative imaging features, then using the model to predict lung cancer and clinical phenotype. Thus, the radiomics can solve the above problems in clinic.The radiomics enterprise can be divide into five distinct processes:(1)image acquisition;(2) image segmentation;(3) feature extraction;(4) generation prediction model;(5) informatics analyses. Radiomics is used to generate a model to discriminate the benign and malignant of lung nodules. According to the information about malignant degree of nodule provided by Lung Image Database Consortium Image Database Resource Initiative(LIDC-IDRI), we divided the nodules of 593 patients into two categories(benign or malignant). To maximize our ability to predict status of nodules, we deliberately designed the training set to contain equal numbers of patients with benign and malignant nodules(200 patients with benign nodules vs 200 patients with malignantnodules). The testing set includes 71 patients with benign nodules and 122 patients with malignant nodules. Here we carry out a radiomic analysis of 200 features quantifying lung tumour image intensity, shape and texture. These features are extracted from 593 patients computed tomography(CT) LIDC-IDRI dataset. Correlation and redundancy between features may reduce the accuracy of the classification, the feature selection method based on minimum Redundancy Maximum Relevance was used to define a feature subset with 15 features. To obtain a higher prediction accuracy, the parameters of SVM are optimized by genetic algorithm. Using support vector machine to train the prediction of benign and malignant of lung nodules model in the training set, and validate in the testing set. The accuracy of prediction of malignant of lung tumor is 86.0% in training set and 76.1% in testing set. As CT imaging of lung tumor is widely used in routine clinical practice, our radiomic classifier will be a valuable tool which can help clinical doctor diagnose the lung cancer.