Respiratory Signals Prediction Based On Particle Swarm Optimization And Back Propagation Neural Networks

In the process of tumor therapy,radiation therapy has been one of main treatment method.Hence,it is important to improve the precision requirement of radiotherapy.The aim of radiation therapy is sending high doses to the tumor area.At the same time,reducing exposure to the surrounding normal tissues and organs.That is killing the tumor cells more accurate and avoiding damage to normal tissues and cells.However,in the process of actual operation,the Radiation Respiratory motion may cause the change of some organs and tissues,such as lung and liver tumors in radiation therapy,which may influence the treatment effect and increase the damage to normal tissues and organs.Hence,it is an essential work to estimate the real-time movement of target in radiation therapy.Respiratory motion may cause the change of some organs and tissues,such as lung and liver tumors in radiation therapy,which may influence the treatment effect and increase the damage to normal tissues and organs.Now following methods were widely used to reduce the effect of respiration: 1).Target area extension could therapy total tumor target.However,it increased the radiation to the normal tissue.2)Breath hold which need patient to hold their breath for a period of time.But some patient couldn't hold their breath.(3)Four dimensional computed tomography(4D-CT)used average respiration of several cycles as real respiration.Hence,it was inaccuracy.(4)respiratory gating.Real-time Position Management? System(PRM)was an import way to compensate the respiratory effect.It used the distance between external markers and camera to estimate real position of tumor.This technology required to obtain real-time position of tumor.Moreover,the therapy equipment had delay time reacting to the operation of human.Hence,it was important to adjust the Multi-Leaf Collimator(MLC)through predict the trajectory of tumor.Now,many scholars,both in China and abroad,had proposed a number of methods to predict respiration.Three methods include linear prediction method,Kalman filter method and neural network method were introduced in this study.Each method had its cons and pros.Someone had verify that the neural network was the most accuracy method in this these methods.BP-NN has been widely used in respiratory motion prediction due to its superior nonlinear fitting capability.However,BP-NN is easy to fall into local minimum.In this study a novel method using PSO to optimize the BP-NN was proposed to avoid its drawbacks and improve prediction accuracy.The preliminary results of 11 patients demonstrate that the mean absolute error reduced from 0.24 to 0.18(25%)and the coefficient correlation increased from 0.82 to 0.86.The proposed method(PSO-NN)could reduce the risk of BP-NN falling into local optimum and has the ability of improving the prediction accuracy of BP-NN method.Besides,the effect of the neuronal number on input-layer(1,3,5,7,9),the neuronal number on hidden-layer(1,2,3,4,5),the initial learning rate(0.001,0.01,0.1,1,10)and the accuracy of loss function(0,0.05,0.1,0.15,0.2)were investigated in this study.The preliminary experiment indicated that the neuronal number on input-layer,the neuronal number on the hidden-layer and the accuracy of loss function were large while the effect of initial learning rate was small.