Automated Segmentation of the Cerebral Ventricles on CT Images

Abstract

Accurate assessment of the volume of cerebral ventricles on computed tomographic (CT) images of the brain is an important and as yet unsolved problem in neuroradiology. Subtle changes in ventricular volume occur early in the development or progression of hydrocephalus, a potentially life-threatening condition that may require urgent surgical treatment. Current subjective assessment of ventricles by neuroradiologists and neurosurgeons has limited accuracy, because of the complex shape of the ventricular system. Comparison of ventricles as depicted on serial imaging studies of the same patient are confounded by differences in the angulations of slices from one study to the next. We are developing an automated system that can segment the cerebral ventricles on axial computed tomographic images of the brain.

Two automated segmentation techniques have been developed and tested. One is based on thresholding and the other on region growing. The results have been compared to a manual segmentation by calculating the similarity index (S). A total of ten cases, each with approximately 20 slices, were tested and a good result (S>0.7) was obtained.