ICA for Ovary Tissue Classification of Perfusion Magnetic Resonance Images

In this study, a method to segment ovary magnetic resonance (MR) images and distinguish healthy tissue from cysts has been described. Through the application of independent component analysis (ICA) to a set of perfusion MR images it was possible to extract the output independent components and their corresponding signal-time curves. After examining and analyzing this result, a polynomial approach was computed to represent the main features of each curve, and automated particular selection of independent components was obtained by applying a Bayesian information criterion able to show the most relevant components. The results shown in this work permit to conclude that the independent components with a step-like signal-time curve allow to distinguish healthy tissue from cysts, thus, giving very promising results for the application of ICA to ovary tissue segmentation of perfusion MR images.