Kernel Feature Analysis for Computer-Aided Detection of Polyps in CT Colonography

A fast kernel feature analysis is presented for 3-dimensional computer-aided detection of colonic polyps on CT colonographic images. The proposed algorithm, called Accelerated Kernel Feature Analysis (AKFA), extracts salient features from a sample of unclassified patterns by use of a kernel method. Unlike other kernel-based feature selection algorithms, AKFA iteratively constructs a linear subspace of a high-dimensional feature space by maximizing a variance condition for the nonlinearly transformed samples. The resulting linear subspace can then be used for defining efficient data representations and pattern classifiers. Numerical experiments based on a feature space, generated from 292 CT colonographic volume scans including 131 polyps on CT colonographic images, showed that AKFA generates concise feature representations, and it yields similar classification performance to that of Kernel Principal Component Analysis (KPCA) whereas AKFA is computationally much faster than KPCA.