Neural Network Prediction of the Trabecular Bone Mechanical Competence Parameter

The mechanical competence parameter (MCP) has been defined to grade the trabecular bone fragility based on the principal component analysis (PCA) evaluated in terms of volume fraction, connectivity, tortuosity and Young modulus of elasticity. Using a set of 83 in vivo distal radius magnetic resonance image samples, an artificial neural network (ANN) has been trained to predict the MCP. After the learning phase, the ANN was able to predict the MCP for 20 new samples with very high accuracy. It is shown that there is a strong correlation (r = 0.99) between the MCP estimated by PCA and ANN techniques. In addition, the Bland-Altman plot provides evidence that the PCA and ANN are reasonably comparable techniques to estimate the MCP.