Radiomic analysis to predict local response in locally advanced pancreatic cancer treated with stereotactic body radiation therapy

Purpose Aim of this study is to assess the ability of contrast-enhanced CT image-based radiomic analysis to predict local response (LR) in a retrospective cohort of patients affected by pancreatic cancer and treated with stereotactic body radiation therapy (SBRT). Secondary aim is to evaluate progression free survival (PFS) and overall survival (OS) at long-term follow-up. Methods Contrast-enhanced-CT images of 37 patients who underwent SBRT were analyzed. Two clinical variables (BED, CTV volume), 27 radiomic features were included. LR was used as the outcome variable to build the predictive model. The Kaplan-Meier method was used to evaluate PFS and OS. Results Three variables were statistically correlated with the LR in the univariate analysis: Intensity Histogram (StdValue feature), Gray Level Cooccurrence Matrix (GLCM25_Correlation feature) and Neighbor Intensity Difference (NID25_Busyness feature). Multivariate model showed GLCM25_Correlation (P = 0.007) and NID25_Busyness (P = 0.03) as 2 independent predictive variables for LR. The odds ratio values of GLCM25_Correlation and NID25_Busyness were 0.07 (95%CI 0.01-0.49) and 8.10 (95%CI 1.20-54.40), respectively. The area under the curve for the multivariate logistic regressive model was 0.851 (95%CI 0.724-0.978). At a median follow-up of 30 months, median PFS was 7 months (95%CI 6-NA); median OS was 11 months (95%CI 10-22 months). Conclusions This analysis identified a radiomic signature that correlates with LR. To confirm these results, prospective studies could identify patient sub-groups with different rates of radiation dose-response to define a more personalized SBRT approach.

Automated summary

The study aimed to evaluate the ability of contrast-enhanced CT image-based radiomic analysis in predicting local response in pancreatic cancer patients treated with SBRT. The analysis identified GLCM25_Correlation and NID25_Busyness as independent predictive variables for LR, with promising predictive value. The findings suggest the potential for a more personalized approach to SBRT in different patient subgroups.