Predictive local receptive fields based respiratory motion tracking for motion-adaptive radiotherapy

Area of Science:

• Medical Physics
• Radiotherapy Technology
• Machine Learning in Medicine

Background:

• Extracranial robotic radiotherapy uses external markers for tumor motion tracking.
• Real-time tumor motion tracking requires prediction models to overcome system latencies.
• Existing methods directly model raw respiratory motion, limiting accuracy.

Purpose of the Study:

• To propose a new prediction algorithm, pLRF-ELM, for enhanced respiratory motion prediction.
• To improve the accuracy of tumor motion compensation in robotic radiotherapy.
• To address the limitations of existing respiratory motion prediction methods.

Main Methods:

• Developed a prediction algorithm based on local receptive fields extreme learning machines (pLRF-ELM).
• pLRF-ELM models higher-level features from raw respiratory motion mapped into a random feature space.
• Evaluated the algorithm using data from 31 patients for two prediction horizons.

Main Results:

• pLRF-ELM demonstrated superior performance compared to existing prediction methods.
• The algorithm accurately predicted respiratory motion for treatment system latencies.
• Abstracted higher-level features effectively approximated nonlinear and non-stationary respiratory motion characteristics.

Conclusions:

• The pLRF-ELM algorithm offers a significant advancement in respiratory motion prediction for radiotherapy.
• Modeling abstracted features provides a more accurate approach to handling complex respiratory motion.
• This method enhances the precision and effectiveness of extracranial robotic radiotherapy.