Finding the difference between periosteal and endocortical bone adaptation by using Artificial Neural Networks

In silico models of bone adaptation successfully simulated in vivo periosteal bone apposition, however, there are instances where these models may have limited success in predicting the new bone formation at endocortical surface. In vivo studies have highlighted that cortical bone surfaces may have differences in their modeling or remodeling responses to mechanical loading. However, the principle which the two cortical surfaces follow in bone adaptation is not very clear. This work accordingly attempts to understand how periosteal and endocortical surfaces accommodate loading-induced new bone formation. A neural network model is used to serve the purpose. A relationship is established to compute new bone thickness as a function of mechanical parameters (normal and shear strains) and non-mechanical parameters (distances from the neutral axis and the centroid) at the two surfaces. Analytical results indicate that two cortical surfaces behave opposite to each other in order to achieve optimal distribution of newly formed bone. The outcomes may be useful in establishing a unifying principle to predict site-specific new bone formation.