Texture analysis of CT images

Abstract
The present study has shown some promise in the use of texture for the extraction of diagnostic information from CT images. A number of features are obtained from abdominal CT scans of the liver using the spatial domain statistical texture analysis methods: SGLDM, GLRLM, and GLDM. This study investigated whether (a) the texture could be used to discriminate among the various tissue types that are inaccessible to human perception and, (b) if so, then what are the most useful feature parameters for such an application? The efficacies of the different methods were evaluated from the consistency of the computed values within a class and from their differences with other classes. The study has demonstrated the use of texture for tissue characterization of CT images. In particular, we have been successful in identifying the onset of disease in liver tissue, which can not be recognized even by trained human observers. Three useful features, namely entropy (H), local homogeneity (L) and grey level distribution (GLD), have been found effective for pattern recognition. The performance of these features has been compared on the basis of statistical significance. The results show that, except for L, (Direction 0/spl deg/) all feature parameters perform equally well and detect early malignancy with a confidence level of above 99%-a finding that will not only help in automation, but more importantly, in early detection of malignancy in the liver.

This publication has 29 references indexed in Scilit: