3637 - Radiomic Signature Predicts Tumor Progression and Survival Outcomes in Pancreatic Cancer Patients Undergoing Stereotactic Body Radiation Therapy (SBRT)
Presenter(s)
Q. Xu1, D. A. S. Toesca2, L. Vitzthum1, A. Jamalian3, E. Schueler4, E. Alkim5, J. R. B. Oo6, D. T. Chang7, G. A. Fisher Jr8, and H. Itakura1; 1Stanford University School of Medicine, Stanford, CA, 2Department of Radiation Oncology, Mayo Clinic, Phoenix, AZ, 3Stanford Center for Biomedical Informatics Research, Palo Alto, CA, 4University of Texas MD Anderson Cancer Center, Houston, TX, 5Stanford University School of Medicine, Palo Alto, CA, 6Stanford University Medical Center, Palo Alto, CA, 7Stanford University, Stanford, CA, 8Department of Medicine, Stanford University School of Medicine, Palo Alto, CA
Purpose/Objective(s): Patients with locally advanced pancreatic cancer (LAPC) face high treatment failure rates, leading to rapid disease progression and poor overall survival (OS). Stereotactic Body Radiation Therapy (SBRT) is an emerging modality for LAPC; however, tools to predict patient outcomes following SBRT remain limited. The purpose of our study was to develop and validate a radiomic signature (RS) from pre-SBRT CT scans to predict rapid tumor progression (=3 months post-SBRT) and overall survival in stage II-III pancreatic cancer patients receiving SBRT following chemotherapy. We hypothesized that a radiomic signature derived from pre-SBRT CT scans predicts rapid tumor progression and poor overall survival in pancreatic cancer patients treated with SBRT in sequence with chemotherapy.
Materials/Methods: This retrospective cohort study included 124 stage II-III pancreatic cancer patients treated with SBRT at a single institution. We collected patient clinical data, including demographics, tumor characteristics, clinical outcome data, pre-SBRT CT imaging, and delineations of tumor regions-of-interest. Using open source software, we extracted 900 radiomic features per tumor and applied LASSO-based feature selection to develop the RS. The RS quantified the likelihood of rapid tumor progression post-SBRT, and predictive performance for OS was assessed alongside ten clinical variables using Cox proportional hazards models. The prediction model was built on the training set (n=74), then validated on the held-out test set (n=50).
Results: Among our cohort (57 men, 67 women; mean age: 67 ± 11 years), the RS based on 43 selected textural features predicted rapid tumor progression with an area under the curve (AUC) of 0.83 in the test set (n=50). High RS was a significant risk factor for mortality: Hazard Ratio (HR) = 2.22 for the first year post-SBRT, increasing to HR = 2.85 beyond one year; non-intensive chemotherapy use increased early mortality risk (HR 1.95, p=0.027); older age was associated with worsened OS beyond one year (HR 1.80, p=0.040).
Conclusion: A CT-derived radiomic signature successfully predicted tumor progression and survival outcomes in pancreatic cancer patients undergoing SBRT. High RS identified patients at greater risk of early distant metastasis and mortality post-SBRT, potentially aiding in personalized treatment strategies. These findings support integrating radiomic-based risk stratification into SBRT treatment planning for LAPC. Further validation in prospective cohorts is warranted.