Predictive value of positioning CT radiomics combined with affected side lung dosimetry parameters for radiation pneumonitis occurrence in patients with breast cancer radiotherapy
Objective To investigate the construction and value of radiation pneumonitis(RP)predic-tive model based on machine learning algorithm.Methods A retrospective analysis was conducted on the clin-ical data in 77 patients with breast cancer receiving radiotherapy and regular follow-up in this hospital from August 2019 to September 2022.The affected side lung was delineated on the localization CT as the area of in-terest and the radiomics features were extracted,meanwhile the affected side lung dosimetric parameters were extracted.After feature screening,the patients were divided into the training set and testing set by a 7∶3 rati-o.The features of positioning CT radiomics were extracted and combined with the dosimetry parameters of the affected side lung,and the model was established by using stochastic gradient descent(SGD)algorithm.The performance of the model was validated by using the area under the receiver operating characteristic(ROC)curve(AUC)and decision curve analysis(DCA).Results Among 77 patients,24 cases developed RP after ra-diotherapy end with an incidence rate of 31.17％.Compared with the patients without RP occurrence,V5,V10,V15,V20,V25,V30 and mean lung dose(MLD)in the patients with RP occurrence were higher,and the differ-ence was statistically significant(P＜0.05).In the training set,36 cases did not develop RP.17 cases devel-oped RP,in the testing set,17 cases did not develop RP and 7 cases developed RP.The affected side lung dosi-metric parameters had no statistical difference between the training set and testing set with and without RP occurrence(P＞0.05).After characteristics screening,the 8 optimal characteristics combinations were finally obtained.The average AUC of SGD model in 50％off cross-validation of the training set was 0.900 and AUC in the test set was 0.882.Conclusion The positioning CT radiomics features combined with dosimetry param-eters of the affected side lung has the good predictive value for RP after breast cancer radiotherapy.
breast cancerradiotherapyradiation pneumonitisradiomicsprediction models
万方数据
维普
