A comparison of gantry-mounted x-ray-based real-time target tracking methods

Area of Science:

• Medical Physics
• Radiation Oncology
• Image-Guided Therapy

Background:

• Modern radiotherapy machines utilize 2D kV imaging systems.
• Integrating 2D-3D inference methods offers potential for real-time 3D tumor tracking.
• Accurate tumor motion assessment is critical for effective radiotherapy delivery.

Purpose of the Study:

• To compare the accuracy of four 2D-3D inference methods for real-time 3D tumor tracking.
• To evaluate the performance of these methods using both inferred (Synchrony) and internal (Calypso) motion traces.
• To investigate factors influencing tracking accuracy, including motion magnitude and correlation.

Main Methods:

• Utilized tumor motion data from the LIGHT SABR trial (Synchrony and Calypso traces) as ground truth.
• Generated 2D kV projections from ground truth trajectories for in silico testing.
• Applied four 2D-3D inference methods: interdimensional correlation, Gaussian PDF, arbitrary-shape PDF, and Kalman filter.
• Quantified tracking errors by comparing inferred 3D positions to ground truth.

Main Results:

• Internal Calypso traces showed larger tracking errors (RMS 0.84-1.25 mm) than inferred Synchrony traces (RMS 0.45-0.68 mm).
• The Gaussian PDF method demonstrated the highest accuracy, followed by Kalman filter, interdimensional correlation, and arbitrary-shape PDF.
• Tracking error positively correlated with motion magnitude across all methods and trace types.
• Interdimensional correlation and breathing periodicity index (BPI) negatively correlated with error for Synchrony traces only.

Conclusions:

• Inferred traces may underestimate kV-based tracking errors due to higher, non-representative interdimensional correlations.
• Internal motion traces (e.g., Calypso) are recommended for accurate kV-based tracking studies.
• The Gaussian PDF method is the most accurate for thoracic/abdominal target tracking.
• Motion magnitude is a significant factor impacting 2D-3D inference accuracy.