Room: AAPM ePoster Library
Purpose: validate the feasibility of radiomic feature extraction from kilovoltage cone-beam CT (kVCBCT) images and to investigate potential application of kVCBCT-based radiomics in predicting radiation-induced pneumonitis (RP).
Methods: thoracic cancer patients treated with radiotherapy during January, 2017 and December, 2019 in our center were retrospectively selected, including 70 patients who reported RP after the treatment. Contoured planning CT (pCT) and kVCBCT acquired at the initial treatment fraction were collected from clinical database for each patient. The dataset was split to training and test groups with the ratio of 4:1. Concordance correlation coefficients (CCC) of 1275 radiomic features extracted from pCT and kVCBCT were calculated. Stable radiomic features (CCC>0.99) were selected to train a logistic regression model that predicted radiation-induced pneumonitis for the 193 patients. Performance of the model was compared with that of another logistic model based on dose statistics (V5Gy, V10Gy, V20Gy, mean dose of lung).
Results: addition to shape and volume, 50 more texture features were found to be consistent between pCT and kVCBCT (CCC>0.99). Considering the well-reported feasibility of pCT-based radiomics, the consistency suggested that kVCBCT images are complementary if not intersubstitutable with pCT for radiomics. Tracking the changing kVCBCT radiomic features during the IGRT course might provide prompt and quantified evaluation of tumor response and OAR risks. AUC values of the two logistic models based on radiomics and dosimetrics were 0.84 and 0.73 respectively. Although RP was directly related to dose, varieties in individual’s radiological sensitivity could undermine the accuracy of dosimetric model, while radiomics model provides additional patient-specific and dynamic indicators.
Conclusion: images are interchangeable with pCT providing a subset of reliable radiomics features. The prognostic value of kVCBCT-based radiomics in predicting RP has been evaluated, which encourages future study on temporal radiomics tracking based on kVCBCT sequence in IGRT.
Funding Support, Disclosures, and Conflict of Interest: Capital's Funds for Health Improvement and Research[2018-4-1027]; Fundamental Research Funds for the Central Universities/Peking University Clinical Medicine Plus X - Young Scholars Project(PKU2020LCXQ019); National Key R&D Program of China(2019YFF01014405); Ministry of Education Science and Technology Development Center[2018A01019]; National Natural Science Foundation of China[11505012,11905150].Corresponding author: Yibao Zhang, email@example.com