Prediction of radiofrequency ablation size of spinal and paraspinal tumors with multiple-needle using machine learning.

The aim of this study was to preliminarily use machine learning and finite element methods to predict the multiple-needle ablation size of spinal and paraspinal tumors and to use genetic algorithms to solve for the optimal ablation parameters under specific ablation size conditions. A two-dimensional two-needle spinal tumor ablation finite element model was created. Its ablation size was analyzed under different approach angles, electrode lengths, electrode diameter, ablation temperatures, and time conditions. Three neural-network models (back propagation (BP), radial basis function (RBF) and convolutional neural-networks (CNN)) were trained separately using the results of the finite element analysis as a dataset (eight inputs and five outputs) and the performance of these neural-network models was compared. The results showed that compared with the RBF and CNN, the BP neural-network has the smallest root mean square error (RMSE) value on the test set (compared with RBF and CNN, the BP neural-network decreased by 55.06% and 56.71%, respectively). This indicated that the BP neural-network has better generalization ability and prediction accuracy compared with RBF and CNN and was more suitable to be used as a machine learning model in this study. Appropriate adjustment of the angle between the needles could effectively control the morphology of the ablation region and avoid damage to the surrounding healthy tissues. Using machine learning and genetic algorithms to predict the size of multi-needle ablation region and optimal ablation parameters could significantly improve research efficiency.