Respiratory motion prediction with surface EMG features

In extracranial robotic radiotherapy, tumour motion is compensated bytracking external and internal surrogates. To compensate system specific time delays,time series prediction of the external optical surrogates is used. We investigate whetherthe prediction accuracy can be increased by expanding the current clinical setupby an accelerometer, a strain belt and a flow sensor. Four previously publishedprediction algorithms are adapted to multivariate inputs - normalized least meansquares (nLMS), wavelet-based least mean squares (wLMS), support vector regression(SVR) and relevance vector machines (RVM) - and evaluated for three differentprediction horizons. The measurement involves 18 subjects and consists of two phases,focusing on long term trends (M1) and breathing artefacts (M2). To select the mostrelevant and least redundant sensors, a sequential forward selection (SFS) method isproposed. Using a multivariate setting, the results show that the clinically used nLMSalgorithm is susceptible to large outliers. In the case of irregular breathing (M2), themean root mean square error (RMSE) of an univariate nLMS algorithm is 0.66 mm andcan be decreased to 0.46 mm by a multivariate RVM model (best algorithm on average).To investigate the full potential of this approach, the optimal sensor combination wasalso estimated on the complete test set. The results indicate that a further decrease inRMSE is possible for RVM (to 0.42 mm). This motivates further research about sensorselection methods. Besides the optical surrogates, the sensors most frequently selectedby the algorithms are the accelerometer and the strain belt. These sensors could beeasily integrated in the current clinical setup and would allow a more precise motioncompensation.