Author: Ming Chao, Hao Guo, Tenzin Kunkyab, Yang Lei, Tian Liu, Kenneth Rosenzweig, Robert Samstein, James Tam, Junyi Xia, Jiahan Zhang 👨🔬
Affiliation: Icahn School of Medicine at Mount Sinai 🌍
Purpose:
The aim of the study is to develop a trade-off prediction model to efficiently guide the treatment planning process for patients with stage III non-small cell lung cancer (NSCLC).
Methods:
We generated 13 volumetric-modulated arc therapy (VMAT) plan variations for each patient in our dataset (n=53); 1 clinical plan and 12 plans incorporating trade-off considerations. The trade-off plans consisted of 3 varying levels (0-2) of sparing priority assigned to each organ-at-risk (OAR), including the esophagus, lungs, heart, and spinal cord. The first 3 principal components of each OAR-plan specific dose-volume histogram (DVH) served as target variable, while 26 morphological features were used as predictors. Patients were randomly split into training set (80%) and test set (20%). A random forest multi-output regression model was trained using the top five morphological features to predict the 3 principal components for each 13 plan-OAR variations. The performance of our trade-off prediction model was compared to a balanced model trained on the clinical plans without trade-off considerations. Evaluation metrics included root-mean-square error (RMSE) and the mean absolute errors (MAE) for DVHs.
Results:
The trade-off prediction model achieved an average RMSE of 5.32 compared to the planned DVHs for all 13 treatment plans, significantly outperforming the balanced model (RMSE = 27.3). The tradeoff model achieved lower MAE for key dose-volume metrics, including cord Dmax (12.5Gy vs 15.5Gy, p<0.01), esophagus Dmax (1.7Gy vs 2.7Gy, p<0.01), left lung V20Gy (7.8% vs 27.5%, p<0.01), right lung V20Gy (7.4% vs 27.8%, p<0.01), and heart V30Gy (10.8% vs 20.9%, p<0.01).
Conclusion:
The proposed multi-output regression model reliably predicts optimal dose objectives and trade-off considerations for NSCLC treatment plans. Incorporating this model into the pre-planning process can aid physicians and planners in their decision-making process by providing feasible trade-off estimations, thereby enhancing the efficiency and quality of the treatment planning workflow.