Research On DVH Prediction Of IMRT Plan Based On ABC-BP Method For Esophageal Cancer

Cancer is one of the main factors threatening the safety and quality of life of human beings,and its treatment is the most concerned focus for the majority of patients and their families,and it is also a hot research direction for scholars today.Esophageal cancer caused by many factors,such as genetics,poor eating habits,bacterial infections and so on,which has become one of the malignant tumors with higher incidence rate in China.Radiation therapy is the main treatment method for esophageal cancer,and intensity modulated radiation therapy(IMRT)is the most widely used radiation therapy technology at present,and the quality of IMRT plan is closely related to the clinical treatment effect for patients.In clinical practice,dose volume histograms(DVH)of target area and organs at risk are usually used to evaluate the feasibility and the degree of superiority and inferiority of the plan.In order to obtain a better DVH,the plan optimization constraints were often adjusted by the planners in the way of artificial trial and error constantly,which consumed a lot of time and energy,and made the patient's waiting time longer before treatment,and increased the possibility of further aggravation of the patient's condition.In this paper,on the basis of fully studying the theory of support vector regression(SVR)and back propagation(BP)neural network learning methods,the correlation models between geometric anatomy and DVH of organs at risk were established for esophageal cancer respectively.Through experiments,it is found that the overall DVH difference range of the four models based on SVR learning method was 2.80% ? 5.85%,among them,the overall DVH difference of left lung model was the smallest,followed by the right lung and spinal cord,and the heart was the largest;the overall DVH difference range of the four models based on BP neural network learning method was 4.71% ? 9.37%,among them,the overall DVH difference of left lung model was the smallest,followed by the right lung and heart,and the spinal cord was the largest;in addition,the overall DVH differences of the left lung,right lung,heart,and spinal cord models based on SVR method were smaller than that of the corresponding organ at risk models based on BP neural network learning method respectively.That is,the model prediction effect based on SVR method was better than BP neural network,but both methods have the problems of easily falling into local minimums,insufficient model stability,and insufficient prediction accuracy.In order to overcome the above problems,a BP neural network learning method based on artificial bee colony(ABC)algorithm optimization was proposed,and it was abbreviated as ABC-BP method,and the DVH prediction model of organs at risk for esophageal cancer was established.Through experiments,it is found that the overall DVH difference range of the four models based on ABC-BP method was 2.80% ? 5.85%,among them,the overall DVH difference of right lung model was the smallest,followed by the left lung and spinal cord,and the heart was the largest.By comparing the SVR,BP neural network,and ABC-BP method,it is found that the overall DVH difference of ABC-BP method which was proposed newly was less than that of the former two.That is,the ABC-BP method was the best in the prediction effect of the organs at risk model,followed by SVR,and the BP network was the worst.In conclusion,SVR,BP neural network and ABC-BP methods were used to establish the correlation model between the geometric anatomy and the DVH of organs at risk for esophageal cancer patients in this paper,and it is showed that the DVH prediction model of organs at risk established by ABC-BP method has better prediction effect by experiments.The establishment of the model can not only provide a good pre-optimization constraints and evaluation criteria of plan quality for planner,but also reduce the optimization time of IMRT plan,waiting time of patients and the possibility of deterioration for esophageal cancer.