Hierarchical finite normal mixtures for post-processing optimization in computed radiography systems

Computed radiography (CR) systems are being increasingly used in the clinical environment as they offer important advantages over traditional radiography, such as requiring lower radiation dosages for comparable image quality due to the inherent linearity in their imaging plate characteristics. In this paper, we study the application of hierarchical finite normal mixtures (HFNM) for modeling the desired parameter settings of the CR system for a particular chosen task, the enhancement of life support lines in chest radiographs. We pose the initial problem as an unsupervised classification problem and use HFNM to discover the structure within the data by using information theoretic criteria and propose ways to improve the robustness of the scheme.