2927 - Optimizing Cardiac Dose Prediction in Left-Sided Breast Cancer Radiotherapy: Clinical Strategies for Identifying Beneficiaries of the Deep Inspiration Breath-Hold Technique
Presenter(s)
S. Y. Chen1, T. Yu1, H. Fang2, H. Jing3, Y. Zhai4, Y. X. Li5, K. Men1, X. Chen5, and S. Wang4; 1Department of Radiation Oncology, National Cancer Center/National Clinical Research Center for Cancer/Cancer Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Beijing, China, 2State Key Laboratory of Molecular Oncology and Department of Radiation Oncology, National Cancer Center/National Clinical Research Center for Cancer/Cancer Hospital, Chinese Academy of Medical Sciences (CAMS) and Peking Union Medical College (PUMC), Beijing, China, 3State Key Laboratory of Molecular Oncology and Department of Radiation Oncology, National Cancer Center/National Clinical Research Center for Cancer/Cancer Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Beijing, China, 4State Key Laboratory of Molecular Oncology and Department of Radiation Oncology, National Cancer Center/ National Clinical Research Center for Cancer/ Cancer Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Beijing, China, 5National Cancer Center/National Clinical Research Center for Cancer/Cancer Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Beijing, China
Purpose/Objective(s): This study evaluated the performance of a convolutional neural network (CNN) model in predicting cardiac doses under free-breathing (FB) and deep inspiration breath-hold (DIBH) conditions for left-sided breast cancer radiotherapy and identifies patients most likely to benefit from DIBH based on cardiac dosimetry.
Materials/Methods: A total of 265 patients with left-sided breast cancer undergoing whole-breast irradiation were included, with 200 retrospectively assigned to the training set and 65 prospectively assigned to the test set. The CNN model incorporated anatomical data, including organ structures and distance-to-target volume maps, to predict three-dimensional dose distributions. Predicted dosimetric parameters were compared with clinical data to assess accuracy. Agreement between model-based and clinical classifications of DIBH benefit was evaluated using kappa statistics.
Results: The CNN model demonstrated high accuracy in predicting cardiac dosimetric parameters, with correlation coefficients ranging from 0.84 to 0.99 for mean dose (Dmean) and D2% in the heart, left anterior descending coronary artery, and ventricles under both FB and DIBH conditions. The model also accurately predicted dose–volume histograms for these structures, with no significant differences between clinical and predicted values. Using a classification approach based on heart Dmean in FB and its reduction via DIBH, the model correctly categorized 90.8% of patients as DIBH beneficiaries, achieving a kappa value of 0.876. When applying a threshold of ?heart Dmean = 1 Gy, the model identified significant DIBH benefits in 63.1% of patients, with 96.9% agreement between predicted and clinical classifications.
Conclusion: The CNN-based model provides an efficient and accurate framework for predicting cardiac dose and identifying patients most likely to benefit from DIBH. Future studies should explore its applicability across broader radiotherapy scenarios and assess its long-term impact on cardiac outcomes.