KEY POINTS
- The model used 182 coronary CT angiography scans to derive probabilistic coronary locations and was evaluated on 60 angiography scans, 34 radiotherapy simulation CTs, and 26 scans from a 0.35-T MR-linac.
- The automated workflow selected 15 anatomically similar angiography templates, deformably registered their coronary contours to the planning image, and incorporated registration, delineation, setup, and cardiac-motion uncertainty into artery-specific probability regions termed habitats.
- Main-branch habitats contained 89.7 ± 11.7% to 98.2 ± 2.6% of manually delineated coronary arteries across modalities. Full-branch habitats derived from the complete angiographic coronary tree contained 92.4 ± 8.9% to 97.6 ± 4.5%.
- When planning-organ-at-risk margins were added to the manual arteries, habitat inclusion fell to 83.1–92.3% on simulation CT and 75.3–87.6% on MR-linac imaging. The largest mean spatial miss was 4.4 ± 4.5 mm for the left anterior descending artery on MR-linac scans.
- Habitat generation took less than four minutes and produced regions representing approximately 2.1–17.4% of whole-heart volume, depending on the artery and imaging modality.
- Across 51 clinical plans, habitat and manually delineated artery-plus-margin doses were strongly correlated: Pearson coefficients were 0.95–0.98 for mean dose and 0.92–0.94 for D0.03cm³, with intraclass correlations of 0.86–0.98.
- In three ultracentral thoracic cases, habitat-guided replanning reduced coronary D0.03cm³ by up to 21.1 Gy and mean dose by up to 18.5 Gy compared with whole-heart-focused plans. All original clinical constraints remained satisfied, although target conformity was slightly reduced and the low-dose spread into the chest wall increased.
CLINICAL TAKEAWAY
The habitat approach could provide a practical coronary-avoidance structure when individual arteries cannot be visualized reliably on simulation CT or MR-linac imaging. Its clinical value remains unproven: registration outliers occurred, only three plans were reoptimized, and no relationship with subsequent cardiac events was evaluated.