Rapid and automated extraction of the fourth ventricle from MR images

Abstract

This paper describes a rapid and automated method that extracts the fourth ventricle from MR brain images in normal and pathological subjects. Anatomical knowledge of fourth ventricle has been incorporated into the method to define a region of interest (ROI), determine intensity thresholds in the histogram of ROI, locate the starting point for the 3D region growing, and extract all parts of the fourth ventricle and aqueduct (even if they are disconnected on images as well as to avoid inclusion of non-ventricular cerebrospinal fluid regions). The method was validated qualitatively on 30 MRI studies with variable parameters. The best overlap metric between a radiology expert and the method was 0.9988 and the worst 0.9621 on 10 quantitative studies. The mean and standard deviation of overlap metric were 0.9806 and 0.0105.