Automatic Recognition of Basal Cisterns on Brain CT

Abstract

Effacement of the basal cisterns (BC) and midline shift (MLS) are two most important features clinicians use to evaluate the severity of brain compression by various pathologies. Because of its complex shape, measuring the compression of the BC is not an easy task and its standardization has not been proposed until recently. Based on this standard method, we develop a method for automatic recognition of the BC on brain CT slices. Hypodense pixels of the brain area on each slice are found with a threshold derived from its own histogram. Hough transform is then applied to find the semicircular band containing largest number of hypodense pixels within the lower-central brain. This area was recognized as the normal or abnormal BC if it fits certain rules derived from human experts. Our system is tested on patient images. We found good inter-rater agreement between the results generated by our system and those evaluated by a board-certified neurosurgeon (kappa = 0.957).