Predictors of Treatment Plan Adaptation In Radiation Therapy for Pancreatic Cancer
Abstract
Purpose
To develop a predictive tool to assess need for treatment plan adaptation.
Methods
Retrospective plans from 23 pancreatic SBRT patients were analyzed for PTV/OAR doses ranked by OAR proximity to PTV. Per-fraction patient data sets were constructed linking daily MR anatomy, adaptive dose metrics (ADM), and clinical decision (adapt plan vs. treat-as-planned). ADM captured anatomical changes by combining organ motion, deformation, overlap index (OI) and dice similarity coefficient (DSC). Conventional center of mass (COM) change was compared to a new proximity index (touching distance, TD) to assess organ motion impact on model predictions. Dosiomic and geometric features were integrated into a unified training set for adaptive modeling, which incorporates regularized logistic-regression baselines with Bayesian hyperparameter (class-weight, λ, α) fine-tuning to identify OARs most likely to cause plan adaptation. Model performance was evaluated with AUROC, sensitivity, specificity, and ODDs ratios (per 1 SD) to indicate features most associated with adaptation likelihood.
Results
Based on 138 image sets, the duodenum dose threshold was typically exceeded at ~1-2mm approach to PTV, stomach and small bowel at ~3-4mm, and large bowel at ≥ 4mm with wide inter-patient variability. A mixed-effects negative-binomial model correlated each 1mm of margin buffer between OAR-PTV to ~7% fewer adapted plans(IRR = 0.93, p = 0.011). Predictive TD model performance (AUROC =0.755) was superior to conventional COM metrics as per sensitivity (0.892 vs 0.501), specificity (0.876 vs 0.633), and prediction power (0.772 vs 0.542) analysis. Absolute TD was a strong indicator of treatment plan adaptation with an ODDs ratio of 3.143(CI = 1.97–5.02, p <0.001), followed by ΔTD, OI, and DSC.
Conclusion
Clinical implementation of predictive adaptive tool can improve treatment efficiency and efficacy for pancreas patients. The proposed framework can be generalized to other treatment sites and image modalities in adaptive radiotherapy.