Pulmonary Structure Classification Using Bayesian Framework
Oral Defence Date:
Professors Okada and Yoon
This thesis presents an innovative Bayesian inference framework to address a medical imaging problem of classifying pulmonary structures with 3D CT scans. Bahlmann et al. previously demonstrated an idea of reducing the 3D topological classification into 2D clustering analysis. Work presented here extends the study of Bahlmann et al. by comprehensive formalization of the original solution with Bayesian approach. Two probabilistic algorithms were derived to solve the classification problem systematically. Furthermore, quantitative evaluation of proposed methods were performed in the context of chest computer-aided diagnosis (CAD) application. Results of quantitative experiments suggest that performance improvement in both accuracy and robustness have been achieved by introducing proposed Bayesian framework into this CAD application.
Topology Classification of 3D Structures, Bayesian Inference, Maximum a posteriori Estimation, Statistical Modeling, Chest CAD, Medical Image Analysis