On content based MRSI retrieval integrating fuzzy descriptors in the wavelet domain

Abstract

Magnetic resonance spectroscopic imaging (MRSI) integrates spectroscopic and imaging methods to acquire spatially localized spectra associated with a specific patient. MRSI is a relatively new imaging entity for clinical applications and gathering relevant data is an expensive task. Therefore, only few small databases might exist for clinical use. However, the rapid advances made in the field of NMR scanning technologies as applied to clinical diagnosis of oncological diseases, may soon lead to the creation of large databases of such images in medical centers and hospitals. Therefore, the need for mining MRSI images will soon emerge. This paper proposes the novel application of a recent method for content based MRSI image retrieval based on investigating fuzzy descriptors in the image and the wavelet domain as an extension of a recent fuzzy descriptor. The description of MRSI images relies on a new descriptor which includes global image features as well as transform domain features, capturing both brightness and texture characteristics at the same time, based on a normalized measure of the MRS spectrum per each image voxel. Image information is extracted using a set of fuzzy approaches applied to image and transform domain. Experiments illustrate the feasibility of the proposed approach using synthetic images derived from benchmark MRS spectra.