AbstractID: 9512 Title: Evaluation of Prediction Methods for Real-Time Tumor Tracking during Treatment Image guided radiotherapy and extra-cranial radiosurgery offers the potential for precise radiation dose delivery to a moving tumor. Recent work has demonstrated how to locate and track the position of a tumor in real-time using diagnostic x-ray imaging to find implanted radio-opaque markers [1]. However, the delivery of a treatment plan that follows the tumor trajectory requires adequate consideration of treatment system latencies, including image acquisition, image processing, communication delays, control system processing, inductance within the motor, and mechanical damping. Furthermore, the imaging dose given over long procedures (radiosurgery) or multiple fractions (radiotherapy) is not insignificant, which means that we must reduce the sampling rate of the imaging system. This study evaluates the ability of various predictive models for reducing tracking error when a real-time tumor tracking system is used to target a moving tumor at a slow imaging rate and with large system latencies. We consider 7 lung tumor cases where the peak to peak motion is greater than 8mm, and compare the tracking error using linear prediction, neural networks, and Kalman filter, against a system which uses no prediction. For imaging rates faster than 10 Hz, we found no benefit in using prediction, but for imaging rates slower than 5 Hz, the prediction methods reduced the tumor localization error. For example, at 3 Hz, the expected error due to system latencies was 1.3mm using prediction vs. 2.3mm without prediction. [1] Shirato, H. et al., Physical Aspects of a Real-Time Tumor-Tracking System for Gated Radiotherapy, Int. J. Radiation Oncology Biol. Phys., Vol. 48, No. 4, pp. 1187–1195, 2000.