Material classification based on Dual-Energy Micro-CT images by the Gaussian mixture model

This study aimed to implement an unsupervised classification method through the Gaussian mixture model to classify different materials using the scatter diagram of the linear attenuation coefficients acquired from dual-energy micro-CT imaging. This method estimates each cluster's distribution parameters and performs classification based on the posterior probability with a pre-determined cluster number. Our studies on dual-energy images of a phantom showed that the distribution of linear attenuation coefficient of different materials on the scatter diagram has a Gaussian distribution, and clusters can be classified using model-based clustering. The result of this classification method is related to the actual materials in the phantom, where a specific cluster represents each material. This classification method can be potentially used when the clusters are overlapped and the material is separated with high accuracy.