1128 - Neural Signals-Based Respiratory Motion Tracking: A Surface Electromyography Study

Purpose/Objective(s): Neural signals-based respiratory motion tracking offers a potential solution to the system latency issue of medical linear accelerators in respiratory motion tracking radiotherapy. However, decoding respiratory-related neural signals from scalp electroencephalography (sEEG) in real-time is challenging. Herein we propose a clinically applicable neural signals-based respiratory motion tracking approach using surface electromyography (sEMG).

Materials/Methods: The neural signal and respiratory motion of fifteen healthy subjects were recorded simultaneously using an sEMG system and a pressure sensor integrated into a stretchy belt. A combination of the gating and the template subtraction methods was used to remove electrocardiogram (ECG) artifacts in the raw sEMG signal. The ECG-removed sEMG signal was smoothed offline to extract respiratory-related neural signal. Cross-correlation analysis was conducted to characterize the time dependencies between the extracted signal and respiratory motion. Using the extracted signal as a reference, the resulting signal from recurrent neural networks (RNNs)-based online smoothing was compared using the mean absolute error (MAE) and root mean square error (RMSE).

Results: The correlation coefficients between the respiratory-related neural signal obtained from offline analysis and respiratory motion consistently exceed 0.90, with the precursor time of the respiratory-related neural signal averaging 319 ms. The resulting precursor time for the first 1-minute interval and the following 9-minute intervals represents no statistical difference. For RNNs-based online smoothing, the obtained MAE and RMSE of the multi-step prediction method with smoothness loss are 0.065 ± 0.025 and 0.083 ± 0.031, respectively.

Conclusion: We have proposed a clinically applicable neural signals-based respiratory motion tracking method using sEMG. The workflow proposed in our study efficiently extracts respiratory-related neural signals with minimal latency, while preserving accuracy. These findings suggest that neural signals-based respiratory motion tracking using sEMG is a promising solution to the system latency issue of medical linear accelerators in cancer radiotherapy.