Application Study Of Radiomics In Individualized Precision Radiotherapy Of Esophageal Cancer

Esophageal cancer is one of the most common malignant tumors in the world.Radiation therapy plays a key role in the clinical treatment of esophageal cancer.Prognostic evaluation of tumors using clinical and radiomics data are effective method for the implementation of individualized precision radiotherapy.This retrospective study explored a new method of automatic extraction of clinical data from 579 esophageal patients,then established a machine learning prediction model for evaluation the two-year survival of esophageal patients after radiotherapy.Machine learning and big data analysis provide a novel method for the implementation of individualized precision radiotherapy.The specific work is as follows:1.Accurate and efficient collection of clinical data is the basis of big data research.We study a new method of automatic extraction of big data in radiotherapy planning,and develop an automated analysis system.The new system analyze the raw data of the treatment plan,which can quickly parse the clinical data of radiotherapy and automatically collect the clinical data of large number patients.This method gets rid of the dependence on the treatment planning system,and more flexible and faster to extract and collect clinical data.2.Radiomics can provide valuable information for individualized precise radiotherapy.At present,the quantification of radiomic features lacks standardization.We built a standardized protocol for extraction radiomic features from images.At the same time,We apply the concept of radiomics to dose data of radiotherapy,and define dosimetric features based on dose data.Those new dosimetric parameters are added to the big data research of radiotherapy.Radiomic features of 579 patients with esophageal cancer were extracted.3.Based on the research of radiomics,the clinical and dosimetric characteristic parameters of radiotherapy are added into the radiomic features.The support vector machines,logistic regression and random forest models were established.We improve the model and the prediction accuracy from the aspects of feature preprocessing,feature selection,and algorithm optimization.The accuracy of the random forest prediction model is the highest,and the classification accuracy rate can reach 90.0%(AUC=0.94).This study uses the radiomic,dosimetric features and machine learning models to evaluate the two-year survival of esophageal patients after radiotherapy,which provides a new paradigm for the study of radiomics in individualized precision radiotherapy.