Estimation of Cortical Bone Strength Using CNN-based Regression Model

Cortical bone is a compact layer that acts as a protective surface and forms an external layer of all bones. With osteoporosis, imbalance between bone formation and bone loss occurs, and this leads to a deterioration of bone microstructure including cortical bone thinning. Therefore, there is a clinical need to estimate and assess bone strength and quality. The detection of bone cortical thickness is still challenging due to the high variance in the speed of sound in the cortical bone. The main aim of this study is to develop an accurate ultrasound method to estimate cortical bone thickness that could be used as a proxy of bone quality by using CNN-based regression models. To achieve this, pulse-echo measurements are performed at multiple ultrasound frequencies and the continuous wavelet transformations (CWT) of the acquired data was used as an input to the CNN. The maximum observed percentages were in 1 mm and 2 mm with an average error of 5.57%, and the minimum error was in group 7 (7mm) with a percentage of 1.6% The preliminary results showed that combination of multi-frequency RF signals has potential to be used for cortical thickness estimation.

Automated summary

The objective of this study is to develop an accurate ultrasound method using CNN-based regression models to estimate cortical bone thickness, which can be used as a proxy for bone quality. The experimental results demonstrate that the combination of multi-frequency RF signals has the potential for cortical thickness estimation.