Automated image-matching technique for comparative diagnosis of the liver on CT examination

When interpreting enhanced computer tomography (CT) images of the upper abdomen, radiologists visually select a set of images of the same anatomical positions from two or more CT image series (i.e., non-enhanced and contrast-enhanced CT images at arterial and delayed phase) to depict and to characterize any abnormalities. The same process is also necessary to create subtraction images by computer. We have developed an automated image selection system using a template-matching technique that allows the recognition of image sets at the same anatomical position from two CT image series. Using the template-matching technique, we compared several anatomical structures in each CT image at the same anatomical position. As the position of the liver may shift according to respiratory movement, not only the shape of the liver but also the gallbladder and other prominent structures included in the CT images were compared to allow appropriate selection of a set of CT images. This novel technique was applied in 11 upper abdominal CT examinations. In CT images with a slice thickness of 7.0 or 7.5 mm, the percentage of image sets selected correctly by the automated procedure was 86.6+/-15.3% per case. In CT images with a slice thickness of 1.25 mm, the percentages of correct selection of image sets by the automated procedure were 79.4+/-12.4% (non-enhanced and arterial-phase CT images) and 86.4+/-10.1% (arterial- and delayed-phase CT images). This automated method is useful for assisting in interpreting CT images and in creating digital subtraction images.