Prediction of bone formation rate of bioceramics using machine learning and image analysis

Population ageing has increased the incidence of osteoporosis, which is mainly treated by using artificial bone. To practically utilise an artificial bone, it is necessary to synthesise bioceramics and evaluate their physical and biological properties, which requires animal experiments. To refine these properties, bioceramics must be evaluated repeatedly, which requires more time, money, and animal sacrifice. In a previous study, a machine learning model was used to predict the bone formation rate as a function of synthesis conditions, physical properties, and implant conditions before performing any animal experiments. However, the model did not consider bioceramic structure, which may be an important predictor of bone formation rate. In this study, we used scanning electron microscope (SEM) images of artificial bones to analyse their structural properties and extract important features for predicting bone formation rate, including grain size and contour, and latent variables were transformed using an autoencoder constructed with SEM images. The features and latent variables were used as inputs for a machine learning model predicting bone formation rate. The coefficient of determination of the proposed model was higher than that of the conventional model, thus confirming that a highly accurate model to predict bone formation rate can be constructed using important features in SEM images. To practically utilize artificial bones, scanning electron microscope (SEM) images of artificial bones are used to analyze their properties and extract important features for predicting bone formation rate, including grain size and contour.