Artificial intelligence surrogate estimated coronary artery dose in lung stereotactic body radiotherapy

An artificial intelligence coronary artery surrogate strongly correlated with manual dose metrics in simulated lung stereotactic body radiotherapy.

KEY POINTS

  • This study evaluated an artificial intelligence-generated left anterior descending coronary artery region as a surrogate for manual left anterior descending coronary artery contours in lung stereotactic body radiotherapy dose assessment.
  • The analysis used deep inspiration breath-hold scans from 49 patients with manual contours and registered 96 lung stereotactic body radiotherapy dose maps, all prescribed 50 Gy in 5 fractions, creating 4704 simulated geometric configurations.
  • The surrogate region contained the manual left anterior descending coronary artery with an average inclusion ratio of 99%, and contained the 3 mm planning risk volume with an average inclusion ratio of 78%.
  • Dose correlation was strong for maximum dose: R² = 0.93 for the manual artery and R² = 0.96 for the 3 mm planning risk volume.
  • Volume receiving at least 10 Gy also correlated well: R² = 0.81 for the manual artery and R² = 0.89 for the 3 mm planning risk volume, while mean heart dose was a poor predictor of artery dose with R² ≈ 0.4.

CLINICAL TAKEAWAY

This artificial intelligence-generated region may offer a scalable way to estimate left anterior descending coronary artery dose in lung stereotactic body radiotherapy, where direct contouring is difficult on typical four-dimensional computed tomography. The main value is enabling larger dosimetric and outcomes studies focused on coronary artery exposure. But this remains a technical validation using registered dose maps and deep inspiration breath-hold anatomy, with robustness on routine four-dimensional computed tomography still under investigation.

SOURCE

Journal of Applied Clinical Medical Physics