Predicting and monitoring response to head and neck cancer radiotherapy using multimodality imaging and radiobiological digital twin simulations.

To predict radiotherapy treatment response for head and neck cancer (HNC) using multimodality imaging and personalized radiobiological modeling.A mechanistic radiobiological model was combined with multi-modality imaging data from diffusion weighted-magnetic resonance imaging and positron emission tomography scans with [F]Fluorodeoxyglucose (FDG) and [F]Fluoromisonidazole (FMISO) tracers to develop personalized treatment response models for human papilloma virus associated HNC patients undergoing chemo-radiotherapy. Models were initialized to incorporate patient-specific imaging and updated to reflect longitudinal measurements of nodal gross tumor volume throughout treatment. Prediction accuracy was assessed based on mean absolute error (MAE) of weekly volume predictions and in predicting locoregional recurrence (LRR) following treatment.Personalized modeling based on pretreatment imaging significantly improved longitudinal volume prediction accuracy and correlation with measurement compared with a generic population model (MAE = 23.4 ± 10.0% vs 24.9 ± 9.0%,= 0.002 on paired-test,= 0.82 vs 0.72). Adding volume measurements from weeks 1 and 2 further improved prediction accuracy for subsequent weeks (MAE = 12.5 ± 8.1%, 10.7 ± 9.9%). When incorporating feedback with longitudinal measurements, penalizing large deviations from pretreatment model parameters using a variational regularization method was necessary to maintain model stability. Model-predicted volumes based on baseline + week-1 information significantly improved LRR prediction compared with week-1 volume data alone (area under the curve, AUC = 0.83 vs 0.77,= 0.03) and was similar to prediction using week-3 volume data (AUC = 0.83 vs 0.85,= non-significant).The proposed approach, which integrates clinical imaging and radiobiological principles, could be a basis to guide pretreatment prescription personalization as well as on-treatment adaptations.