Parametric and Non-Parametric Nodule Models : Design and Evaluation

Lung nodule modeling quality defines the success of lung nodule detection. This paper presents a novel method for generating lung nodules using variational level sets to obtain the shape properties of real nodules to form an average model template per nodule type. The texture information used for filling the nodules is based on a devised approach that uses the probability density of the radial distance of each nodule to obtain the maximum and minimum Hounsfield density (HU). There are two main categories that lung nodule models fall within; parametric and non-parametric. The performance of the new nodule templates will be evaluated during the detection step and compared with the use of parametric templates and another non-parametric Active Appearance model to explain the advantages and/or disadvantages of using parametric vs. non-parametric models as well as which variation of nonparametric template design, i.e., shape based or shape-texture based yields better results in the overall detection process.