2707 - Delta Radiomic Analysis during Radiotherapy Potentially Predicts Oncologic Outcomes in Locally Advanced Vulvar Cancer
Presenter(s)
A. Alzibdeh1, B. Hammadeh2, R. J. Alnajjar3, M. Abd Al-Raheem3, R. Mheidat1, A. Al matairi4, M. S. Qamber1, H. M. Almasri3, A. Al-Omari3, and F. J. Abuhijla1; 1Department of Radiation Oncology, King Hussein Cancer Center, Amman, Jordan, 2Faculty of Medicine, Al-Balqa' Applied University, Salt, Jordan, 3King Hussein Cancer Center, Amman, Jordan, 4Faculty of Medicine, University of Jordan, Amman, Jordan
Purpose/Objective(s): Response to radiotherapy in locally advanced vulvar cancer may predict outcomes. We hypothesized that delta radiomic features of gross tumor (GTV) acquired during definitive radiotherapy may serve as prognostic markers for local control (LC), regional control (RC), and overall survival (OS).
Materials/Methods: Using an automated pipeline implemented utilizing PyRadiomics, radiomic features were extracted from GTV on CT simulation scans from patients with locally advanced unresectable vulvar cancer before and after the first phase of radiotherapy. Delta radiomics were calculated and z-score normalized. A two-step feature selection was applied: logistic regression with L1 regularization combined with Recursive Feature Elimination with Cross-Validation (RFECV) identified candidate predictors, followed by bootstrapping (1,000 iterations) to assess feature stability. Delta radiomics were correlated with outcomes using univariable and multivariable Cox regressions, with stepwise backward elimination and bootstrapping to assess predictor stability.
Results: In 21 patients, 42 contoured images were evaluated. A total of 107 quantitative features were extracted. After a median follow-up of 50 months, 2-year LC, RC and OS were 56.2%, 65.7% and 55.9%, respectively. For LC, 2 delta features were selected. Among these, multivariable analyses revealed that change in the original Neighborhood Gray Tone Difference Matrix (NGTDM) Busyness remained the sole robust predictor of LC, but did not reach statistical significance (p = 0.182). Regarding RC, change in the original Gray Level Run Length Matrix (GLRLM) Run Variance and change in the original Shape Maximum 3D Diameter were selected. Multivariable analyses showed that higher values of these features were significantly associated with increased risk of regional failure (HR 2.81, 95% CI: 1.15–6.87, p = 0.02 and HR 3.89, 95% CI: 1.31–11.54, p = 0.01, respectively). For OS, five delta features were selected. Among these, multivariable analyses showed that the two most influential predictors were change in the original Shape Surface-to-Volume Ratio (HR 3.37, 95% CI: 0.82–13.86, p = 0.09) and change in the original Gray Level Size Zone Matrix (GLSZM) Gray Level Variance (HR 2.30, 95% CI: 0.76–6.94, p = 0.14).
Conclusion: Our results suggest that, during radiotherapy, specific delta radiomic features—particularly those reflecting alterations in tumor texture and shape—are associated with clinical outcomes in locally advanced vulvar cancer. These findings warrant further validation in larger cohorts and support the exploration of machine learning algorithms for enhanced predictive modeling using these features.