Font Size: 

Background: To establish a treatment planning to predict therapeutic parameters accurately and avoid other healthy tissue damage by radiofrequency ablation (RFA) in spinal tumor.  Materials and Methods: According different therapeutic parameters (voltages 10-30V and heating-up time 120-1200s) and structures of electrode, the ablation results were achieved by finite element method to build a database. For the purpose of predicting therapeutic parameters in the treatment planning through the database, we chose three modes to compare the best one, including back propagation (BP) neural network, general regression neural network (GRNN) and support vector machine (SVM).  Results: Small tumor (maximum diameter<30.0mm) could be ablated by single needle electrode, and large tumor (maximum diameter≥30.0mm) should be applied by multi needle electrode. Besides, acquired 383 sets of data were regarded as database by the simulation. Compared with the other two models, SVM model is the best one to be applied in the treatment planning, in which error rates(0 and 0.33%) were the smallest and correlation coefficients(1.000 and 0.999) were the closest to 1. To verify whether the ablation results are correct, we have compared them to the clinical treatment results and previous studies.  Conclusions: This treatment planning by SVM is accurate to predict voltage and heating-up time, and it has higher correlation coefficient and few errors.