Texture Analysis as a Tool for Tissue Characterization in Clinical MRI

Abstract

Magnetic resonance imaging (MRI) is a valuable tool for medical diagnosis, as it is a non-invasive technique that allows superior visualisation of soft tissues. Because of the vast growth of the acquired information from medical images the development of new computer-aided diagnosis (CAD) systems has become increasingly important. The application of texture analysis (TA) in the diagnostic interpretation of MR images has become a rapidly expanding field of research. The goal of this thesis was to test the feasibility of texture analysis methods in diagnostic radiology. In this dissertation, texture analysis was applied to three different clinical materials. This study investigates whether the texture could be used to discriminate breast cancer and visible and non-visible changes in brain MRI of mild traumatic brain injuries and multiple sclerosis patients and, if so, which is the optimal texture analysis method for these applications. This study showed that TA could provide a quantitative method to aid radiologists in the detection and classification of pathological findings. A case-specific selection of the texture parameters from histogram-, co-occurrence-, run-length- and wavelet- based methods would be the optimal solution for the evaluated clinical applications. However, larger study samples are needed to further validate these findings. Another conclusion was that the texture analysis process should be simplified considerably and implemented in other CAD systems to be considered for clinical use in the future.