A novel feature extraction scheme for visualisation of 3D anatomical structures

The purpose of this research is to analyse various algorithms for feature extraction in CT images and thereby developing an efficient scheme for feature extraction that supports effective visualisation of three-dimensional (3D) bony structures. An extensive study has been made on the single-scale and multi-scale spatial domain feature extraction algorithms. In order to overcome the issues associated with spatial domain feature extraction algorithms, a frequency domain-based scheme involving curvelet transform has been introduced. The significant features are extracted by designing a filter based on the statistical parameters of the curvelet coefficients across multiple scales. The slices with significant features are stacked along the z-plane to generate a point cloud. The effectiveness of the proposed curvelet-based algorithm is proved by comparing it with other methods based on Feature Similarity Index Measure (FSIM). Feature similarity was found to be between 85% and 90% for the curvelet-based feature extraction scheme.