基于机器学习的肺癌放疗后放射性肺炎预测研究OA

Objective To establish a prediction model of radiation pneumonitis based on the clinical dosimetric data of lung cancer patients before radiotherapy,and to explore the weights of characteristic factors affecting radiation pneumonitis.Methods The clinical dosimetric data of 126 lung cancer patients before radiotherapy were selected.Three machine learning methods,namely logistic regression analysis,support vector machine and random forest,were used as base classifiers.Radiation pneumonia ensemble learning(RPE)prediction model was established through the voting method.Results The accuracy of the RPE model in predicting radiation pneumonitis was 71.16%,the sensitivity was 68.31%,and the specificity was 73.98%.The receiver operating characteristic curve and the area under curve of RPE reached up to 0.798±0.082.In the weight analysis of the influencing factors of radiation pneumonitis,the weight of bilateral lung V20(the volume percentage of organs receiving at least 20 Gy dose,and so on for others)was the highest at 0.2423,followed by the weight of bilateral lung V40 at 0.1624.Conclusion The RPE model established based on the clinical dosimetric dataset of radiotherapy patients can effectively improve the prediction accuracy of radiation pneumonitis compared with three traditional base classifiers.Moreover,this model can provide a guiding basis in the design of radiotherapy plans and improve the quality of life of patients.