Texture analysis in radionuclide tomographic liver imaging.

Abstract

In clinical trials assessing the value of radionuclide tomographic liver imaging there has been no significant increase in diagnostic accuracy when conventional views have been viewed with tomographic images. In an effort to increase the diagnostic value of tomography, we have applied some simple methods of texture analysis to both phantom and patient tomograms in order to quantify different uptake patterns. Three algorithms were used which look at the differences between a pixel and its near neighbours. In phantom experiments, the variation of image texture with image noise was quantified. Accurate values of texture were obtained for images of the phantom containing various sizes of cold spots. Detection limits were then calculated for the different sizes of cold spots. It was found that 1.5 cm cold spots could be detected with a collection time of approximately 25 min, the usual collection time for livers being 20 min. The theoretical detection limits were checked experimentally and found to agree well with experimental values. A preliminary clinical survey was made applying texture analysis to eight images from normal livers and eight from abnormal (metastatic) livers. Although the sample was small, good separation between normals and abnormals was obtained.