Pretreatment [18F] FDG-PET texture analysis to predict local response of pancreatic cancer to radiotherapy.
375 Background: Accurate assessment of radiographic response following radiotherapy (RT) for pancreatic adenocarcinoma is challenging. Morphologic and textural features of FDG-PET have been shown to correlate with pathologic response and clinical outcomes in other solid tumors (PMID 23204495). The goal of this study was to develop a predictive algorithm derived from textural features of PET scans to predict response to RT. Methods: With IRB approval, we reviewed 10 patients with locally advanced pancreatic cancer treated with stereotactic body radiation therapy (25-30 Gy in 5 daily fractions). 18FDG-PET/CT scans were obtained 2 weeks pre-RT and 6 weeks post-RT. Pre-RT PET/CT images were deformably registered to the RT planning CT. Tumor volumes of interest were divided into (4.8mm)^3 subvolumes and characterized by mean SUV uptake, RT dose and comprehensive texture analysis. These pre-RT variables were correlated to post-RT mean SUV to identify potential predictors of treatment response. Response prediction was modeled by logistic regression with the Lasso algorithm and validated by 10-fold cross-validation. Model performance was assessed using cross-validated area under the receiver operating characteristic curves (AUC). Results: Mean uptake, RT dose and 6 texture features (energy, correlation, variance, sum mean, cluster tendency, and inverse variance) on pre-RT PET scans were significant in predicting treatment response (AUC 0.85). Within this model, each of the above noted variables was predictive of post-RT response (p<.05). Conclusions: Subvolume-based metabolic and texture features of pre-treatment PET scans were predictive of response following RT. Studies are ongoing to further correlate these variables to RECIST and pathologic response. This should serve as a useful model to help direct response-driven adaptive radiotherapy in patients with locally advanced pancreatic cancer.