Feasibility test of prospective tracking of respiration signals for breathing compensated or gated radiotherapy

Motion compensation radiotherapy systems are currently available. Signals representing moving targets can be obtained from either real-time X-ray images of implanted markers near the target or external respiration sensors. However, a time lag exists between the motion signal acquisition and action execution (moving the robotic arm). This time lag has been estimated as 0.3 s, and a typical respiratory velocity of 2∼4 mm/sec results in a systematic position lag between 1 and 2 mm. In this study, the feasibility of prospective tracking of respiration signal based on autoregressive (AR) model was tested to compensate this time lag. Free breathing motion signals were acquired by an electromagnetic 3D position measurement system. The software for signal acquisition and AR model computation was developed as a visual C++ application. The signals were obtained at a frequency of 3 Hz. Original respiratory signal of T seconds (variable) was used to compute the AR model parameters. The model order was also variable. The model was updated every T seconds. The same model was used for T seconds while the input data was updated with the actual signals. The predicted positions were computed and the errors to the actual positions were computed. The position errors allowing for time lag were computed through shifting the motion signal by one time unit (∼0.33 seconds). Different methods to compute the AR model coefficients were tested, among which the Burg´s method has shown slightly better error reduction capability. In a typical test case, the mean absolute error was reduced from 0.8 mm to 0.25 mm for a respiratory signal of 1 min.