2022 World Conference on Lung Cancer (ePosters)-EP05.01-015. Validate Radiomics Features and XGBoost Model in Radiation Pneumonitis (RP) Prediction in Patients with Primary Lung Cancer: A MultiCenter Study

EP05.01-015. Validate Radiomics Features and XGBoost Model in Radiation Pneumonitis (RP) Prediction in Patients with Primary Lung Cancer: A MultiCenter Study

Back to course

Pdf Summary

This study aimed to validate the use of radiomics features and an XGBoost model in predicting radiation pneumonitis (RP) in Chinese patients. RP is a common side effect of radiotherapy in lung cancer patients. Previous studies have shown that baseline lung condition and radiation dosimetric factors are risk factors for RP. The study used radiomics features extracted from CT scans and dosiomics features from radiation dose maps to develop and validate a predictive model using the XGBoost method. The study included a total of 513 patients from four centers in China. The primary endpoint was grade 2 RP. The predictive power of the model was assessed using the area under the receiver operating characteristic curve.<br /><br />The results showed that the XGBoost model had acceptable performance, with an AUC of 0.750 for the test set of Center #1. Baseline lung radiomics and dosiomics were found to be significant factors for RP in both the testing and validation datasets. Dosiomics appeared to be more important than radiomics in predicting RP. The best performance was observed in the XGBoost model that combined all clinical factors, dosimetric factors, and baseline lung radiomics.<br /><br />However, it was observed that the CT imaging modality used (different types of CT scans and diagnostic criteria) affected the generalization ability of the predictive model. The incidence of RP varied among different centers, and the performance of the model was worse in centers with different CT types.<br /><br />In conclusion, the study validated the use of radiomics features and an XGBoost model in predicting RP in Chinese patients. Dosiomics was found to be more important than radiomics in the prediction of RP. However, the generalization ability of the model was affected by the CT imaging modality. Further research is needed to optimize the model for different CT types and diagnostic criteria.

Asset Subtitle

Jiali Liu

Meta Tag

Speaker Jiali Liu

Topic Locally Advanced Non-small Cell Lung Cancer - Chemoradiotherapy and Radiotherapy

Keywords

radiomics features

XGBoost model

radiation pneumonitis

Chinese patients

lung cancer

baseline lung condition

radiation dosimetric factors

CT scans

dosiomics features

predictive model