基于小波分解和神经网络的呼吸运动预测算法

Respiratory motion prediction based on wavelet and neural network

目的放射治疗是胸腹部肿瘤治疗的常用手段,但呼吸等运动大大影响了放射治疗的准确性,因此精确的呼吸运动定位和预测对肿瘤治疗很有必要。相关预测方法缺乏对系统长延迟预测的研究,本文提出一种用小波分解结合Elman神经网络的算法(wavelet Elman network,WEN)预测呼吸运动。方法采用光学定位系统采集数据,对数据进行简单的预处理,再利用小波分解压缩数据,训练Elman神经网络,最后进行神经网络的预测。预测结果和真实值对比,绘制误差曲线,计算均方根误差,并与其他主流算法对比,验证算法的可行性。结果 WEN算法在短延迟预测中表现一般,但当延迟达1000 ms时,WEN算法的均方根误差平均为1.6164 mm,比临床中使用的线性预测低32.9%。结论通过实验验证了基于小波分解和Elman神经网络的呼吸运动预测算法,在长延迟时表现较好,证明了本算法的正确性及可行性。

Objective One of the common treatments for cancer of the chest and abdomen is radiation therapy, yet respiratory movement reduces the quality of radiation therapy. Therefore, the precise positioning and prediction of respiratory movement are essential in radiation therapy. An algorithm for predicting respiratory movement is proposed based on wavelet and Elman network since there is a lack of research on long delays in prediction. Methods First, we collected data by optical positioning system, pre-processed the data, then compressed data by wavelet,trained Elman network, and predicted the neural network. With the comparison of predictions and actual values, we drew error curve, calculated the root mean square error, and compared with the other major algorithms to validate the feasibility. Results The performance of WEN in short delays was not very well ,when the delay came to 1000 ms ,the average of WEN＇ s RMSE （root mean square error） was 1. 6164 ram, 32.9% less than linear prediction used in clinical. Conclusions The experiments demonstrated that the respiratory motion prediction based on wavelet and Elman network performed well in the long delays, and all the results demonstrated the validity and feasibility of the algorithm.