The extraction of the best SGLD texture features in the ultrasound B-scan images of cancered stomach coats

Abstract

SGLD (spatial gray level dependence) matrices are used to analyze the B-scan images of 23 samples of normal stomach coats and 14 samples of cancerous stomach coats. According to these matrices, the values of eight texture features of each sample image are computed. Two groups of conditional frequency distributions are obtained. On the basis of these distributions, the authors evaluated the quality, which reflects the error probability in discriminating between pattern classes of all the features. By comparing the measurements of the quality, the authors select from these features the most effective ones in discriminating between a normal stomach and a cancerous stomach. The evaluation methods include normal distribution hypothesis testing, and T testing. The result of the experiments indicates that the selected texture features can be applied to an automatic diagnosis system in the near future.<>