ABSTRACT
This chapter introduces a novel framework for the classification of lung nodules by modeling the nodules' appearance features using a novel higher-order Markov-Gibbs random field (MGRF) in addition to geometric features. The appearance feature is modeled using an MGRF that is used to relate the joint probability of the nodule appearance and the energy of repeated patterns in the three-dimensional (3D) scans in order to describe the spatial inhomogeneities in the lung nodule. The proposed framework overcomes the limitations through the integration of a novel appearance feature using seventh-order higher-order MGRF that takes into account 3D spatial interaction between the nodule's voxels and geometrical features extracted from the segmented lung nodule with the deep autoencoder to achieve high classification accuracy. Describing the visual appearance features using the MGRF model will distinguish between benign and malignant nodules showing high distinctive features.
