3399 - Value of Peritumoral Radiomics in Predicting Local Recurrence-Free Survival in Esophageal Squamous Cell Carcinoma Receiving Definitive (Chemo)Radiotherapy

J. Li¹, L. Zhao², and J. Gong³; ¹Department of Radiation Oncology, Xijing Hospital, Air Force Medical University, Xi’an, China., China, ²Department of Radiation Oncology, Xijing Hospital, Fourth Military Medical University, Xian, Shaanxi, China, ³Department of Radiation Oncology, Xijing Hospital, Air Force Medical University, Xian, China

Purpose/Objective(s): CT or PET radiomics has been used in prognosis prediction for esophageal squamous cell carcinoma (ESCC) receiving definitive (chemo)radiotherapy (CRT). However, most of these studies primarily focus on exploring the radiomics of the core region of the tumor, disregarding the crucial information present in the tumor surroundings, reflecting the tumor microenvironment. The aim of this study was to investigate the additional value of CT and PET peritumoral radiomics in predicting local recurrence-free survival (LRFS) following CRT in ESCC. Materials/

Methods: 234 patients ESCC with contrast-enhanced CT and 18F-fluorodeoxyglucose PET images before CRT in our hospital were retrospectively collected, which were randomly assigned to the training (n = 187) and validation sets (n = 47). Tumor region of interest (ROI) was delineated by two experienced radiation oncologists. The peritumor ROI area was manually annotated including the adjacent tissues around the esophagus and lymph nodes, making sure to exclude the trachea, aorta, and vertebrae. Extract radiomic features from intratumoral and peritumoral regions of PET and CT images, respectively, and construct radiomic signatures. The median of radiomic signature was used as the cutoff to divide patients into high and low-risk groups. Kaplan-Meier analyzed the LRFS of patients in different risk groups. And, predictive models for LRFS were established by integrating intratumoral and peritumoral radiomics features in the training set and tested in the validation set.
Results: A total of 14 intratumoral and 9 peritumoral radiomic features based on CT, as well as 20 intratumoral and 28 peritumoral radiomic features based on PET, were selected as the most valuable predictors of LRFS. These features were chosen during the training set to construct the radiomic signature. Whether it is based on PET or CT images, radiomic signatures from intratumoral or peritumoral regions can stratify patients into different risk groups, with significantly higher LRFS in the low-risk group compared to the high-risk group (P<0.05). This result was also validated in the validation set. The hybrid radiomics models, incorporating both intratumoral and peritumoral radiomic features, outperformed models based on intratumoral features alone in in both PET and CT images. In the training set, the C-index for the hybrid models was 0.735 for CT and 0.767 for PET, compared to 0.719 for CT and 0.721 for PET using intratumoral features only. In the validation set, the C-index for the hybrid models was 0.697 for CT and 0.752 for PET, while it was 0.676 for CT and 0.729 for PET using intratumoral features alone.
Conclusion: In addition to intratumoral radiomic features, peritumoral radiomic features also provide valuable information about the tumor. These features capture the intrinsic characteristics of the surrounding tissue and can complement the intratumoral features. By combining both types of features, we can enhance the accuracy of prognostic predictions for ESCC.