A real-time IGRT method using a Kalman filter framework to extract 3D positions from 2D projections.

Area of Science:

• Medical Physics
• Radiotherapy
• Image-guided therapy

Background:

• Accurate 3D prostate motion estimation is crucial for effective radiotherapy.
• Current methods may require initial learning periods, increasing patient dose.
• Real-time motion tracking enhances treatment precision.

Purpose of the Study:

• To develop and evaluate a Kalman filter (KF) framework for real-time 3D prostate motion estimation from 2D kV images.
• To assess the KF method's accuracy, robustness to noise, and computational efficiency compared to existing techniques.
• To reduce patient imaging dose by eliminating the need for an initial learning phase.

Main Methods:

• A KF framework was implemented for real-time 3D motion estimation from simulated 2D prostate projections.
• Adaptive estimation of the noise covariance matrix and initialization using population covariance were employed.
• The method was validated in silico using patient motion data and simulated VMAT conditions with added noise.

Main Results:

• The KF method achieved sub-millimeter accuracy (0.3-0.4 mm 3D RMSE) without noise.
• With simulated noise, the KF method demonstrated 3D RMSEs of 1.1 ± 0.1 mm.
• This performance was superior to the Gaussian PDF method, which yielded a 3D RMSE of 2 ± 0.1 mm with noise.

Conclusions:

• The developed KF method provides a fast, accurate, and robust solution for real-time 2D to 3D prostate motion estimation.
• The approach effectively handles the random-walk nature of prostate motion and is highly resilient to measurement noise.
• This technique holds significant potential for improving image-guided radiotherapy for prostate cancer patients.