ABSTRACT
Nuclear and cell density features have been measured in 22 cases of glioblastoma divided into two groups according to their survival periods, i.e. less than 12 months or more than 12. The results have demonstrated that the logarithmic transformation of the following features show up statistically significant differences (p less than 0.05): mean of the logarithm of nuclear area, standard deviation of the logarithm of perimeter and standard deviation of the logarithm of the roundness factor. The standard deviation of the roundness factor has been shown to be another parameter with statistically significant differences between the two groups. Forward stepwise discriminant analysis was adopted in order to identify the quantitative features which contributed most significantly to discriminating between the two groups. The results suggested a combination of two features, i.e. the standard deviation of the logarithm of the roundness factor and the mean of the logarithm of the roundness factor. The comparison between actual and predicted categories showed 68.18% agreement: 7 out of 22 cases were allocated incorrectly by the computer. However, when a classification probability threshold was adopted, the 7 incorrectly allocated cases assumed an "intermediate" position between the two groups, in agreement with their survival.
