Research on developing a predictive model for radiation pneumonitis risk based on radiomics and multiparameters

Abstract

Abstract Objective A Nomogram model was constructed by combined pre-treatment CT radiomics, clinical characteristics, and lung dosimetry data of patients with non-small cell lung cancer,and to explore its predictive value of radiation pneumonitis. Methods A retrospective analysis was conducted on 104 non-surgical NSCLC patients who underwent chest intensity modulated radiation therapy(IMRT) at our center from January 2013 to December 2017. Intratumoral and peritumoral radiomics models were established using pre-radiotherapy CT images, and logistic regression was used to screen for the best clinical and dosimetric parameters. A combined Nomogram model was established by the above parameters, and receiver operating characteristic curve (ROC) analysis and area under the curve (AUC) was performed to estimate its predictive efficacy of radiation pneumonia. Results Among 104 patients, 59 cases developed radiopneumonia and 45 cases did not suffer from radiopneumonia within 6 months after radiotherapy, which were divided into 73 cases in the training set and 31 cases in the validation set. The AUC values of the intratumoral radiomics group model in the training and validation sets were 0.871 (95%CI 0.771~0.938) and 0.719 (95%CI 0.400~0.952), respectively, and the predictive efficacy was better than that of the peritumoral radiomics group model [0.798 (95%CI 0.629~0.921) and 0.714 (95%CI 0.500 ~0.857)]. Multifactorial regression analysis showed that patients' age, smoking, and pre-radiotherapy lymphocyte ratio were associated with radiation pneumonitis (P<0.05); physical dosimetric parameters of MLD and lung V20Gy were associated with radiation pneumonitis (P<0.05).The Nomogram model constructed by the intratumor radiomics model combined with clinical and dosimetric parameters had AUC values of 0.928 (95%CI 0.879~0.966) and 0.765 (95%CI 0.692~0.831) in the training and validation sets, respectively.It has the best prediction efficacy. Conclusion The Nomogram model based on the intratumoral radiomics features of pre-radiotherapy CT images, patient’s age, smoking, and pre-radiotherapy lymphocyte ratio combined with MLD and lung V20Gy has a better predictive accuracy for radiation pneumonitis in NSCLC, and it can be used as a quantitative model for the prediction of radiation pneumonitis in patients undergoing radiotherapy for NSCLC.