Neural Network Based Bone Density Estimation from the Ultrasound Waveforms Inside Cancellous Bone Derived by FDTD Simulations

Quantitative ultrasound techniques for bone assessment now attract strong research attentions because of their non-invasiveness, portability, and the low diagnosis expense. The analysis of the ultrasonic signals propagating along the cancellous bone is important because it strongly reflects the bone quality. However, it is difficult to analytically understand the wave behavior because the cancellous bone has complexed porous structure. Therefore, the neural network-based approaches were used for the estimation of the bone density. The waveforms propagating inside the cancellous bone were derived by the FDTD simulation. As a result, the neural network-based method showed a potential to estimate the bone density better than the traditional method.