Numerical simulation and machine learning based analysis of the ultrasonic waveform propagating in bone tissue

Abstract

Since bone has a complex structure, it is difficult to analytically understand the behavior of the ultrasound although it is useful for bone quality diagnosis. Our group, therefore, had been working on simulating ultrasound propagation inside bone models. In this paper, the results of the neural network based bone analysis using waveforms derived by the 3-D elastic FDTD (finite-difference time domain) simulation will be presented as well as its basis and some representative results of the FDTD method applied for the bone assessment. Since the FDTD simulation only requires the 3-D geometry of the model and the acoustic parameters (density, speed of longitudinal wave and shear wave) of the media, it has been very useful for evaluating each effect of a certain acoustic parameter or the bone geometry such as bone density, respectively, in addition to the purpose of visually understanding the wave behavior inside the model including the propagation path. Moreover, thanks to the recent powerful computational resource, it is now realized to prepare a huge number of waveforms for machine learning by using FDTD method. As a result, it was shown that the neural network can estimate the bone density better than the traditional method. (Grant: JSPS KAKENHI 16K01431.)