Detection of temporal bone abnormalities using hybrid wavelet Support Vector Machine classification

Abstract

Support vector machine (SVM) is a machine learning algorithm, which learns to perform the classification task through a supervised learning procedure, based on pre-classified data examples. SVM uses kernel mapping to map the non-linear data in input space to a high-dimensional feature space where the data is linearly separable. A hybrid wavelet kernel construction for support vector machine is introduced in this paper. The construction involves a multi-dimensional sinc wavelet function together with one of the conventional kernel functions. We show that the hybrid kernel is an admissible support vector (SV) kernel satisfying Mercerpsilas theorem. The hybrid kernels thus constructed are used for the automated detection of temporal bone abnormalities. From high resolution computed tomography (HRCT) images features are extracted and fed to the learning machine for classification. Hybrid kernels provide better classification of the signal points in the mapped feature space, compared to conventional kernels. The experimental results denote promising generalization performance with the hybrid kernels.