Respiratory prediction algorithm based on adaptive neuro-fuzzy inference system / 国际生物医学工程杂志

ABSTRACT

Objective To study an accurate prediction algorithm for irregular respiratory movements,effectively compensate for the time delay of radiotherapy systems,so as to improve the target accuracy of imageguided tracking or gated radiotherapy for thoracic and abdominal tumors.Methods A prediction algorithm based on an adaptive neuro-fuzzy inference system (ANFIS) was proposed to precisely predict irregular respiratory motion.The ANFIS model structure adopted the position and velocity of respiratory motion as input parameters to construct an N×N fuzzy set that combines position and velocity,and to establish a training set through historical data.In the prediction,if the position or velocity in the latest input signal was out of the range of the training set,the position or velocity would be adjusted accordingly and treated as the input parameter of the ANFIS model for prediction.The ANFIS was evaluated using 20 cases of irregular respiratory clinical data from CyberKnife-treated thoracic and abdominal tumors patients.The accuracy of the four typical prediction algorithms,including ANFIS,neural network (NN),support vector machine (SVM),and CyberKnife Synchrony,was compared by retrospective offline analysis.Results The ANFIS algorithm showed better performance than NN,SVM and Synchrony in normalized root mean square error (nRMSE),the maximum error (Max),and the number of errors greater than 1 mm.Conclusions The accuracy and robustness of the ANFIS algorithm are superior to NN,SVM and CyberKnife Synchrony.The ANFIS can better predict irregular respiratory signals.