KEY POINTS
- The retrospective study included 150 patients and 2,400 fractions of left-sided post-mastectomy chest-wall and regional nodal VMAT delivered with DIBH to 42.56 Gy in 16 fractions.
- Model development used 30 patients, independent testing used treatment-day CBCT-derived synthetic CT dose recalculation, and an additional 100-patient deployment cohort was used to assess workflow scalability.
- Respiratory features included breath-hold amplitude deviation, a stability index and baseline drift, combined with planning CT geometry and baseline plan dosimetry.
- Mean heart dose was predicted relatively well. External validation produced a mean absolute error of 0.58 cGy per fraction and R² of 0.641.
- LAD D0.03cc was substantially harder to predict, with MAE 24.14 cGy and R² 0.435, consistent with the sensitivity of a small coronary structure to steep dose gradients and contour-transfer uncertainty.
- In the external recalculation cohort, treatment-day mean heart dose differed from planning by an average +2.76 cGy per fraction, while LAD D0.03cc differed by +75.3 cGy per fraction, with considerably greater dispersion.
- Respiratory stability and baseline drift were more influential predictors than simple breath-hold amplitude, suggesting that remaining within the gating window does not guarantee identical cardiac geometry.
- The framework remains a decision-support concept. RPM reflects external surface motion rather than internal heart position, synthetic CT provides an estimated rather than directly measured delivered dose, target coverage was not modeled simultaneously, and no validated clinical threshold exists for triggering intervention.
CLINICAL TAKEAWAY
DIBH quality may be better characterized by how stable the breath-hold is, not simply whether the respiratory trace remains inside the gating window. A respiratory-based model could help identify fractions for closer image review, but it should not independently trigger coaching changes or adaptation, particularly for LAD dose.