Automated segmentation and retrieval system for CT head images

Abstract

In this paper, automatic segmentation and retrieval of medical images are presented. For the segmentation, different unsupervised clustering techniques are employed to partition the Computed Tomography (CT) brain images into three regions, which are the abnormalities, cerebrospinal fluids (CSF) and brain matters. The novel segmentation method proposed is a dual level segmentation approach. The first level segmentation, which purpose is to acquire abnormal regions, uses the combination of fuzzy c-means (FCM) and k-means clustering. The second level segmentation performs either the expectation-maximization (EM) technique or the modified FCM with population-diameter independent (PDI) to segment the remaining intracranial area into CSF and brain matters. The system automatically determines which algorithm to be utilized in order to produce optimum results. The retrieval of the medical images is based on keywords such as "no abnormal region", "abnormal region(s) adjacent to the skull" and "abnormal region(s) not adjacent to the skull". Medical data from collaborating hospital are experimented and promising results are observed. © 2009 Springer-Verlag.