Deep learning based time series analysis for breathing phase prediction in phase-gated proton therapy

Area of Science:

• Medical Physics
• Radiotherapy
• Artificial Intelligence in Medicine

Background:

• Phase gating is crucial for mitigating tumor motion in spot-scanned particle therapy (SSPT).
• Current real-time phase prediction methods struggle with patient-specific breathing variability and system latencies, leading to suboptimal radiotherapy delivery.
• Improved phase prediction accuracy is needed for efficient and precise radiation delivery, minimizing manual intervention.

Purpose of the Study:

• To enhance the accuracy of breathing phase prediction for radiotherapy.
• To leverage deep learning (DL)-based time series forecasting for improved phase prediction.
• To enable more accurate dose delivery in spot-scanned particle therapy (SSPT).

Main Methods:

• Retrospective breathing waveforms from 69 proton therapy patients were analyzed.
• Deep learning models, including Long Short-Term Memory (LSTM) and Temporal Fusion Transformer (TFT), were trained to predict breathing phases.
• Model performance was evaluated using mean squared error, phase binning accuracy, and timing deviation, with Bayesian optimization for hyperparameter tuning.

Main Results:

• The curated dataset comprised over 260,000 seconds of breathing waveform data for training, validation, and testing.
• LSTM and TFT models demonstrated significant improvements over a commercial algorithm, increasing phase prediction accuracy by nearly 20%.
• Timing deviations were reduced across all breathing regularity levels, indicating more precise radiation delivery timing.

Conclusions:

• Deep learning-based time series forecasting offers a substantial improvement in breathing phase prediction accuracy compared to current clinical methods.
• This approach provides a more precise and reliable method for phase-gated radiation delivery in particle therapy.
• The findings suggest a promising pathway toward optimizing radiotherapy efficiency and accuracy.