Prediction of the skull's acoustic parameters in transcranial focused ultrasound based on neural network

Abstract

In this paper, multilayer perceptron neural networks (MLP neural networks) were used to predict the monkey skull's acoustic parameters (density and sound velocity) for transcranial focused ultrasound (tcFUS) therapy based on the computed tomography (CT) of the monkey skull. Levenberg-Marquardt algorithm was applied for the training of the neural networks with 168 input learning materials. The predicted results of eight samples demonstrated the effectiveness of the proposed method for accurately predicting the monkey skull's acoustic parameters with the maximum absolute error and the maximum relative error for density of 0.0541 kg/m3 and 0.00355% respectively and for sound velocity of 0.167 m/s and 0.00542% respectively. Further comparisons of the two dimensional (2D) transcranial focused ultrasound simulations with the predicted and the hypothetically real acoustic parameters of the monkey skull showed that the maximum absolute error and the maximum relative error of the acoustic field were 2.3 × 10−4 and 0.00161% respectively. Additionally, at the main distribution area of the acoustic field, the relative error was smaller. All of the numerical analyses of the density error, sound velocity error and acoustic field error were responsible for supporting the proposed method suitable for precisely predicting the monkey skull's acoustic parameters.