Hierarchical iterative Bayesian approach to automatic recognition of biological viruses in electron microscope images

The diagnosis of biological viruses appearing in electron microscope images is currently based on time consuming visual examination by highly trained and experienced medical specialists. To reduce the diagnosis time and to de-skill the diagnosis task by allowing the use of non-specialist medical staff, an automatic virus recognition method is presented. The method is based on a hierarchical approach to decompose the multi-category multi-feature classification problem into a set of two-category classification sub-problems, with each classification sub-problem solved based on an iterative Bayesian approach. Probability of misclassification is minimised by searching an optimum set of features selected from virus spectra and projecting them to the first principal component axis. The proposed method and the reliability are described and demonstrated using four different biological viruses