A multiple model approach to respiratory motion prediction for real-time IGRT

Area of Science:

• Medical Physics
• Radiotherapy
• Image Guidance

Background:

• Respiration causes significant tumour motion in thoracic and abdominal radiotherapy.
• Real-time image-guided radiotherapy (IGRT) aims to adapt radiation delivery to this motion.
• Time lag in tumour position acquisition leads to beam positioning errors and affects dose coverage.

Purpose of the Study:

• To develop and evaluate a predictive algorithm for respiratory-induced tumour motion.
• To address the time lag issue in real-time IGRT.
• To propose a confidence interval (CI) criterion for evaluating prediction performance.

Main Methods:

• Utilized an interacting multiple model (IMM) filter combining constant velocity (CV) and constant acceleration (CA) models.
• Employed Kalman filters for local models, with the IMM filter integrating their predictions.
• Proposed a confidence interval (CI) criterion for performance evaluation.

Main Results:

• The IMM filter consistently outperformed individual Kalman filters (CV or CA models).
• Prediction errors for the IMM filter were within 2.1 mm (95% CI) at 0.4 s prediction time.
• Errors were within 3.6 mm (95% CI) at 0.6 s prediction time, with minimal difference between 5 and 10 Hz sampling rates.

Conclusions:

• The IMM filter effectively predicts respiratory tumour motion for IGRT.
• The proposed 95% CI criterion is a clinically relevant measure for assessing prediction performance and setting margins.
• The IMM filter offers a promising solution for reducing motion-related errors in radiotherapy.