Abstract:
HTML
Purpose: Early CT scanning of stroke patients is performed to determine whether hemorrhage is present as this influences the subsequent treatment. We describe a computer algorithm to automatically detect acute hemorrhage on CT images and present an initial assessment of its accuracy.
Methods and Materials: CT images of the head were acquired using a standard helical protocol reconstructed with a 3mm slice thickness. Fully automatic image analysis was performed using a priori knowledge of the characteristics of the normal head and hemorrhage. The primary basis for identification of hemorrhage was its Hounsfield number. First, the skull was segmented based on attenuation and connectivity and the cranial cavity located. The cerebrospinal fluid (CSF) was similarly segmented. Normal structures which might have similar attenuation to blood were excluded on the basis of position, shape and local morphology. In particular, punctate calcification (where voxel attenuation may fall within the range of fresh blood) was identified by morphology. The dural folds and dural venous sinuses were located by their shape and their position relative to the skull. Choroid plexus and pineal gland were differentiated from hemorrhage by their position in the CSF spaces. The performance of the system was tested on images from 30 stroke patients who were sequentially referred for CT imaging to look for hemorrhage.
Results: The algorithm correctly reported 29 of the 30 CT datasets (97%). One small thalamic haemorrhage (3mm in size, visible on a single slice) was missed. No normal structures were incorrectly attributed to haemorrhage.
Conclusion: The algorithm presented accurately identifies hemorrhage on CT images in the majority of cases. Extensive hemorrhage which may require urgent surgical intervention is reliably detected.
Questions about this event email: RichardHodgson@btinternet.com
Hodgson, R,
Automatic Detection of Hemorrhage in CT Imaging of Stroke. Radiological Society of North America 2003 Scientific Assembly and Annual Meeting, November 30 - December 5, 2003 ,Chicago IL.
http://archive.rsna.org/2003/3103669.html
