Incorporating and quantifying deformable image registration uncertainties in dose accumulation: a feasibility study...

Objective. Accurate dose accumulation relies on deformable image registration (DIR) to track dose across multiple images. However, DIR introduces uncertainties that can impact cumulative dose distributions. In this study, we present a probabilistic framework that explicitly incorporates DIR uncertainties into dose accumulation, translating them into clinically relevant metrics via dose-volume histogram (DVH) bands. As a clinical use case, we applied this framework to a small patient cohort to demonstrate its feasibility and explore its potential for improving adaptive proton therapy evaluation.Approach. A previously validated deep learning model was used to quantify DIR-related uncertainties in five head-and-neck cancer patients (157 daily cone beam computed tomographys). Synthetic CTs were generated for each fraction and deformably registered to the planning CT using 100 probabilistic deformation vector fields per fraction. These were used to warp the daily dose distributions and generate probabilistic cumulative doses visualized as DVH bands. Two adaptive workflows were compared: (i)Triggered_APT-Offline replanning triggered when deemed clinically necessary. (ii)DAPT-Daily online adaptive proton therapy with full daily replanning.Main results. DIR uncertainties were successfully integrated into dose accumulation and translated into interpretable metrics. Across the analyzed cases, DAPT consistently improved target coverage and OAR sparing compared to Triggered_APT(e.g. D98% improvement up to 4 GyRBE), even when dose accumulation uncertainties are explicitly modeled.Significance. Our findings demonstrate the feasibility and relevance of uncertainty-aware dose accumulation. The framework offers an interpretable way to visualize DIR-related uncertainty and could support the evaluation and refinement of DIR-dependent adaptive workflows.