1991 年 40 巻 1 号 p. 31-34
Using phantom images, the adaptation of texture analysis was evaluated in analyzing the scintigrams. Prior to analyse the scintigrams, image data acquisition and preprocessing to degrade statistical noise of pixel counts must be performed. Feature parameters which represent RI distribution should be selected on texture analysis of the scintigrams. Using the feature parameters extracted from probability of pixel counts, the more ununiformity became in RI distribution, the less energy became, but mean, variance, skew and entropy became larger values. When discriminant analysis was done using the textural features, ratio of correlation was 0.998, and accuracy of classification on phantom images which were difficult to be classified visually became 58-83%. The textural features was estimated to be useful quantitative data, when the scintigrams was read visually.