Respiratory Instability-driven Cardiac Dose Deviation in Postmastectomy Deep Inspiration Breath-hold Volumetric Modulated Arc Therapy (DIBH VMAT): Development and External Validation of a Respiratory-informed Predictive Framework.

Abstract

AimsDeep inspiration breath-hold (DIBH) reduces cardiac exposure during left-sided breast radiotherapy; however, treatment-day cardiac dose may differ from the planned dose because of variations in respiratory reproducibility and anatomy. The aim of this study was to develop and validate a framework for predicting fraction-level cardiac dose variation using routinely acquired respiratory, anatomical, and dosimetric information.Materials and methodsA retrospective cohort of 150 patients treated with postmastectomy DIBH VMAT (volumetric modulated arc therapy) was analyzed. Models were developed using a 30-patient development cohort with patient-wise-nested cross-validation, externally validated in 20 independent patients using cone-beam CT-derived synthetic CT dose recalculation, and applied to a 100-patient deployment cohort. Separate models predicted deviation from planned per-fraction mean heart dose (ΔMHD) and left anterior descending coronary artery dose (ΔLADCA D0.03 cc) using treatment-day real-time position management (RPM)--derived respiratory features, planning computed tomography (CT)-derived geometric variables, and plan-derived dosimetric parameters. SHAP-based analysis was performed to identify influential predictors.ResultsThe final Gradient Boosting model for ΔMHD achieved an internal validation mean absolute error (MAE) of 0.541 ± 0.035 centigray (cGy) and coefficient of determination (R²) of 0.642 ± 0.069. The Ridge Regression model for left anterior descending coronary artery (ΔLADCA) achieved an MAE of 23.413 ± 1.617 cGy and R² of 0.413 ± 0.082. External validation demonstrated good agreement for ΔMHD (MAE 0.58 cGy, R² = 0.641), whereas ΔLADCA showed lower performance (MAE 24.14 cGy, R² = 0.435). Stability index and baseline drift were key predictors.ConclusionTreatment-day-respiratory characteristics combined with anatomical and dosimetric variables can estimate fraction-level cardiac dose variation during DIBH radiotherapy. The framework may support verification-based clinical review, with further validation required before implementation.