Study On Application Of Radiomics And Artificial Intelligence To Predict Radiation Pneumonitis

Part 1 Study on predicting radiation pneumonitis based on radiomics technologyObjective To investigate the value of radiomics technology in predicting the occurrence of RP,and identify the radiomics features related to the occurrence of radiation pneumonitis.Methods Retrospective analysis of 86 patients with stage Ⅲ non-small cell lung cancer who received IMRT.The radiation pneumonitis was graded by follow-up imaging data and clinical information.The lung was used as the volume of interest.The radiomics features were extracted from planning CT images.Analyzing the radiomics features,clinical and dosimetric parameters associated with radiation pneumonitis.Using the support vector machine to construct the model,and the model prediction performance was evaluated by the five-fold verification method.Results A total of 1029 radiomics features were extracted from CT images and 5 features were selected by ANOVA and LASSO.Two validation sets show difference between adopting radiomics features only and incorporating clinical parameters,dosimetric c and radiomics features(AUC,0.67 and 0.71,respectively).Conclusions The radiomics model constructed by planning CT images of lung cancer patients has the potential to predict the occurrence of radiation pneumonitis.Addition of clinical and dosimetric parameters,superior model performance relative to radiomics features is produced.Part 2 Prediction of the radiation pneumonitis by deep learningObjective The CT images of patients with locally advanced non-small cell lung cancer and limited-stage small cell lung cancer were analyzed,and the feasibility of using deep learning methods to establish a prediction model of radiation pneumonitis was discussed.Methods Retrospective and prospective collection of 303(retrospective)and 35(prospective)patients with locally advanced non-small cell lung cancer and localized small cell lung cancer,followed by radiographic data and clinical information to classify and collect radiation pneumonitis.It locates CT images and performs preprocessing such as lung selection,image resampling,and image cutting and compression.The Res Ne Xt50 network is used to establish a prediction model of radiation pneumonitis.The retrospective data set uses 5-fold cross-validation,500 epochs,and then uses retrospective data as training set,and the prospective data is used as verification set to use the same method to verify the model prediction ability,establish the ROC curve,and use the area under curve AUC to evaluate the prediction model ability.Results The predictive model based on retrospective data is a 5-fold cross-validation AUC of 0.71,0.79,0.73,0.74,0.74,and the average AUC is 0.75;using prospective data as a validation set,AUC is 0.66,0.65,0.69,0.67,0.65,and the average AUC is 0.66.Conclusions The deep learning method has the ability to predict radiation pneumonitis,and it is necessary to expand the prospective data to further improve the model.