Deep learning generates free-breathing body contours from CT scouts for DIBH radiotherapy

Deep learning generated free-breathing body contours from biplanar CT scout images with high accuracy (Dice 0.98±0.02, 95% Hausdorff 4.2±2.9 mm), potentially eliminating extra CT for DIBH setup, but prospective validation needed.

• Retrospective study of 173 thoracic CT-simulation studies (121 training, 26 validation, 26 test).
• Larger, more symmetric torso shapes associated with better contour prediction.
• Single-center design; requires prospective validation in clinical setup and gating workflows.

RadPigeon summaries are original and for information only. They are not clinical advice.