Chongqing University Cancer Hospital Shapingba, Chongqing
X. Yang1, J. Liu2, B. Feng1, H. Luo1, L. Chen1, L. Tan1, and F. Jin1; 1Radiation Physics Center, Chongqing University Cancer Hospital, Chongqing, China, 2Department of Oncology,Mianyang Central Hospital, School of Medicine,University of Electronic Science and Technology, Mianyang, China
Purpose/Objective(s):This study aims to utilize deep learning techniques to comprehensively integrate pre-treatment CT images, clinical features, dosimetric parameters, radiomic features, and 3D dose distribution data for constructing a dual-center predictive model for the efficacy of concurrent chemoradiotherapy in esophageal squamous cell carcinoma.Materials/
Methods: A retrospective analysis included data from 369 esophageal squamous cell carcinoma patients who underwent concurrent chemoradiotherapy between January 2018 and November 2022 across two medical centers. Dosimetric factors were calculated from dose-volume histogram, such as average and maximum doses for the planning target volume. Clinical features included gender, age, smoking and alcohol history, and clinical staging. A total of 1349 radiomic features were extracted, and cluster analysis was employed to identify radiomic features correlated with treatment efficacy. The deep learning algorithm underwent end-to-end training, integrating the multiple data sources to establish a complex predictive model. Patients were randomly divided into training (90%) and validation (10%) sets, and ten-fold cross-validation was performed to assess model stability. Sensitivity, area under the curve (AUC), and other metrics were utilized to evaluate model performance. Results: A total of 112 highly correlated radiomic features were selected for cluster analysis. The model demonstrated a validation set accuracy of 0.804, sensitivity of 0.798 (95% confidence interval [CI] [0.235, 0.986]), and specificity of 0.821 (95% CI [0.349, 0.994]). The AUC ranged from 0.76 to 0.92 in five-fold cross-validation. Conclusion: Through the successful application of deep learning techniques, this study proposes a comprehensive approach integrating multiple data sources for predicting the efficacy of concurrent chemoradiotherapy in esophageal squamous cell carcinoma. The model not only exhibits excellent predictive performance, but also providing new insights for personalized treatment strategies. The successful application of this method holds the potential to offer more precise support for individualized treatment decisions in clinical practice, contributing to the potential improvement of patient prognosis.
