Correlation and variation of a multi-modal sensor setup for respiratory motion prediction and correlation

Purpose: In robotic radiosurgery, tumor movements are compensated by tracking external optical surrogates. These surrogates are used to compensate for time delays (prediction) and to calculate the internal fiducial position (correlation). We aim to increase the accuracy and robustness by using a multi‐modal sensor approach including different physiological sensors. We evaluate the correlation coefficient of a strain belt, acceleration and air flow sensor with respect to external standard optical sensors and an internal landmark in the liver tracked using 3D ultrasound and evaluate the variance for a measurement over 20 minutes.

Methods: We recorded sensor data from 6 subjects (5 male/1 female). All sensors have been synchronized and downsampled to the optical sampling rate (fs = 47 Hz) or in case of internal correlation to the ultrasound sampling rate (fs = 17 Hz). Pearson's correlation coefficient r was calculated for each minute, discarding the first and last minute.

Results: The mean (standard deviation) external correlation coefficients over all subjects and time periods were obtained as: 0.88 (0.036) for strain, 0.75 (0.024) for flow and 0.73 (0.052) for acceleration with respect to an optical marker on the chest. The mean (standard deviation) internal correlation coefficients are: 0.81 (0.045) for strain, 0.76 (0.041) for flow, 0.58 (0.088) for acceleration and 0.80 (0.057) for the optical marker on the chest with respect to the internal landmark.

Conclusion: This study indicates that apart from the optical markers, strain and flow data show the best correlation to external and internal motion and seem to be promising for increasing the prediction and correlation accuracy as well as robustness. Among the investigated sensors, the strain data have the lowest standard deviation for internal and external correlation, being even lower than the standard deviation of the optical chest marker.