Modeling Circulating Lymphocyte Dose during Lung Proton Therapy Using AI-Based Pulmonary Vasculature Segmentation and 4D Blood-Flow
Abstract
Purpose
Radiation-induced lymphopenia (RiL) is linked to impaired anti-tumor immunity and poorer outcomes in thoracic malignancies. Reducing dose exposure to circulating lymphocytes (CLs) could therefore mitigate RiL. Because Intensity Modulated Proton Therapy (IMPT) and proton arc delivery differ in beam geometry, scanning patterns, and delivery time, they may alter blood exposure and CL depletion. We evaluated their mechanistic differences during lung proton radiotherapy.
Methods
We evaluated retrospectively 10 lung cancer cases treated with IMPT (PTV: 182.3 cm³±47.1). An alternative proton arc plan was generated using identical dose constraints. Patient-specific pulmonary vasculature was extracted using an AI-based segmentation model trained on thoracic CT, yielding individualized arterial and venous maps. To enable such assessment, we developed a 4D dosimetric blood-flow model that integrates beam delivery dynamics (gantry rotation/spot timing) with cardiopulmonary circulation to estimate cumulative CL dose. We computed irradiated-blood-volume metrics and estimated CL depletion. This work introduces the “effective CL-sparing metric” as a novel metric that links dose distribution and dose rate, depleted CL, and the interplay between treatment time and systemic circulation period into a single dosimetrically meaningful endpoint
Results
Proton arc exhibited increased temporal overlap with peak pulmonary blood-flow compared to IMPT (15.2% vs 18.3%). IMPT showed higher heart–lung shunt dose due to prolonged spot-scanning near the mediastinum, while proton arc reduced recirculation passes per CL. The effective CL-sparing metric was reduced by 39.5% for PBSArc compared with IMPT, indicating a lower sparing potential.
Conclusion
Proton arc results in substantially greater irradiation of CL than IMPT. The AI-based segmentation of vasculature combined with 4D hemodynamics enables mechanistic evaluation of CL dose in proton lung therapy. The proposed metric provides a unified endpoint for RiL-oriented treatment optimization.