Patient-Specific Modeling for Optimizing Pulse Timing In Personalized Ultra-Fractionated Stereotactic Adaptive Radiotherapy (PULSAR) for Non–Small Cell Lung Cancer
Abstract
Purpose
Personalized Ultra-Fractionated Stereotactic Adaptive Radiotherapy (PULSAR) delivers high-dose radiation pulses separated by intervals of several weeks, enabling tumor response–guided adaptation, reduced toxicity, and potential synergy with immunotherapy. However, the optimal timing between radiation pulses remains undefined: intervals that are too short may underestimate tumor response, whereas overly long intervals risk confounding response with tumor regrowth. We developed a patient-specific mechanistic model to characterize tumor dynamics during PULSAR using sparse longitudinal tumor volume data.
Methods
Thirty-seven non–small cell lung cancer patients were analyzed, including 24 treated with PULSAR and 13 with conventional stereotactic body radiotherapy (SBRT). A minimal two-compartment ordinary differential equation (ODE) model was constructed, representing active tumor cells (T) and inactive (radiation-damaged) tumor cells (I). Active tumor growth followed logistic kinetics and radiation-induced cell kill was modeled using the linear–quadratic formalism, while inactive cells decayed with a clearance rate. The individualized radiosensitivity parameter (α) was estimated using the Levenberg–Marquardt algorithm by minimizing the squared residuals between simulated and observed tumor volumes, while all other model parameters were fixed. Patients were classified as responders or non-responders based on final tumor volume relative to baseline. Model performance, response discrimination, and post-pulse shrinkage timing were evaluated.
Results
The model accurately reproduced tumor volume dynamics, with normalized root mean square errors (NRMSE)of 0.120 ± 0.058 for SBRT and 0.144 ± 0.085 for PULSAR. Model-derived predictions differentiated responders from non-responders with an AUC of 0.95 (SBRT) and 0.87 (PULSAR). Simulations demonstrated that, except for extremely radiosensitive or radioresistant patients, maximum tumor shrinkage following each PULSAR pulse occurred at 24 ± 10 days, compared with 35 ± 13 days for SBRT.
Conclusion
This patient-specific model estimates an individualized optimal interval between pulses, supporting adaptive evaluation of tumor response and radiosensitivity during PULSAR.