Real-time prediction of respiratory motion based on local regression methods

Area of Science:

• Medical Physics
• Radiotherapy Technology
• Computational Biology

Background:

• Conformal radiation dose delivery to small, localized targets is advancing.
• Real-time target motion tracking and prediction are crucial for image-guided radiotherapy (IGRT) systems to manage latency.
• Minimizing sampling rates in IGRT is desirable to reduce patient imaging dose.

Purpose of the Study:

• To develop and evaluate a real-time prediction method for respiratory motion, specifically impacting lung tumors.
• To address the challenges in predicting complex and variable breathing patterns.

Main Methods:

• Proposed a prediction method based on local regression.
• Utilized an augmented state space to capture system dynamics.
• Employed local regression in the augmented space, leveraging the semi-periodicity of respiratory motion.
• Incorporated local weighting adjustment to account for fading temporal correlations.

Main Results:

• The proposed local regression method demonstrated reduced root mean square error (RMSE) in predicting tumor motion across various imaging rates and latency lengths.
• The method showed particular effectiveness for longer prediction lengths.
• Outperformed traditional methods like linear prediction, neural networks, and Kalman filtering in prediction accuracy for the evaluated dataset.

Conclusions:

• The developed local regression-based prediction method offers improved accuracy for real-time respiratory motion management in radiotherapy.
• This technique has the potential to enhance the precision of conformal dose delivery for lung tumors.
• The findings support the utility of advanced prediction models for overcoming IGRT system limitations.