Evaluation of Markerless Spine Position Monitoring for Auto Beam Hold Based on Triggered Imaging
Abstract
Purpose
To evaluate the performance of a 2D markerless spine position monitoring technique for Auto Beam Hold based on kilovoltage triggered imaging.
Methods
Kilovoltage triggered images were acquired at 0.33 and 2 fps on a Varian TrueBeam during delivery of 10 MV FFF VMAT arcs. Experiments were performed using a CIRS Dynamic Thorax Motion Phantom and in-house developed 3D-printed anthropomorphic thoracic and lumbar spine phantoms. The 3D‑printed thoracic phantom could be configured with or without arms, while the lumbar phantom included removable rings to simulate different patient sizes (left-right diameters of ~25–45 cm). For each dataset, a single vertebra was monitored using a template-based tracking algorithm. Imaging parameters were evaluated across a range of energies (80–120 kVp) and mAs settings (5, 10, and 15 mAs). Geometric accuracy was assessed with the phantoms positioned at the isocenter, with positional offsets (0.5–2.5 mm in individual directions and combinations), and during stepwise (parallel and perpendicular to the kV detector) and gradual motion. In addition, triggered imaging data acquired during spine SBRT of 16 patients were analyzed.
Results
For the lumbar phantom, 80 kVp/5 mAs were found to be adequate for 25 cm diameter, 90 kVp/5 mAs for 33 cm, 90 kVp/10 mAs for 38 cm, and 120 kVp/10 mAs for 45 cm. For the thoracic phantom, 80 kVp/5 mAs sufficed without arms (~26-28 cm diameter), and 90 kVp/5 mAs with arms attached (~41 cm diameter). Across static and dynamic phantom experiments, the mean squared localization error in the isoplane was <0.3 mm. For patient data, successful vertebra localization was achieved in 97.5% of 5146 images; failures mainly occurred when imaging through the shoulders or arms.
Conclusion
The evaluated algorithm accurately and robustly localized vertebrae in phantom experiments and the majority of patient data, achieving sub‑millimeter accuracy suitable for clinical spine position monitoring.