基于支持向量回归的呼吸运动预测技术的研究

Study on predicting respiratory motion with support vector regression

胸腹部肿瘤放疗中,由于呼吸运动的影响需要对靶区进行实时跟踪以保证放疗精度,并通过预测来补偿系统延时。本研究提出一种基于支持向量回归的呼吸运动预测方法,该方法先选取一段呼吸运动序列进行训练得到回归模型,当有新的呼吸序列时,根据训练模型计算输出。并在此基础上,动态更新训练集,使模型在线更新,实现精确在线支持向量回归。实验中对7例呼吸运动样本数据分别用离线模型和在线模型进行训练并预测,平均绝对误差分别为0.42 mm和0.30 mm。在线精确支持向量回归能更准确刻画呼吸运动轨迹,拟合结果精度高,满足实际应用中的需求。

The target is usually tracked in real time at thoracic and abdominal radiotherapy due to the effect of respiratory motion,the prediction is necessary to compensate the system latency. A prediction method based on support vector regression（ SVR） was proposed,a part of historical data for training was selected,and then the output was calculated according to the training model when there was a new sequence. Furthermore,the training set would be dynamically updated and the accurate online support vector regression model was achieved. The experiment selected seven respiratory motion data; the model was trained by on-line and off-line method,then prediction was carried out. The mean absolute error was 0. 42 mm,0. 30 mm,respectively. The respiratory motion is accurately described by the online accurate support vector regression,and the results with high precision can satisfy practical application.