基于分离有限状态模型的呼吸预测算法

Respiratory prediction algorithm based on a separated finite state model

由呼吸运动导致的肿瘤位置变化严重影响放射治疗精度,需要准确预测呼吸位置以实现系统的实时在线误差补偿。该文提出了一种基于分离有限状态模型的呼吸预测算法,将呼吸基线与起伏相分离,并分别使用局部加速度恒定的Kalman滤波和改进的有限状态模型进行预测。改进的有限状态模型将呼吸数据按照邻域特征分为线性、非线性和不规则3种状态,根据每种状态的特点用不同的模型进行预测。对25组患者的实际呼吸数据进行预测,并和传统预测算法进行对比,结果表明:分离有限状态模型呼吸预测算法的均方误差和最大误差明显下降,尤其对于有明显基线漂移的呼吸状态,该算法的预测误差几乎不受影响。

Respiratory tumor motion seriously affects the accuracy of radiation therapy, so predictions of the respiratory motion in advance will allow real-time compensation. A respiratory motion prediction algorithm is developed based on a separated finite state model. The baseline and the up-and-clown motion are separated and then predic- ted using a Kalman filter for the baseline and an improved finite state model for the up-and-down motion. The respiration data is represen- ted by linear, nonlinear and irregular states in the finite state model. Different models are used for the predictions according to the charac- teristics of each state. This algorithm is used to predict the respira- tory motion for 25 patients. The results show that the average root mean square error and the maximum error over all the patients are significantly lower than for traditional prediction algorithms. Espe- cially for the respiratory motion with baseline drift, the algorithm er- ror is almost unaffected.