The computer-assisted microscope analysis of Feulgen-stained nuclei linked to a supervised learning algorithm as an aid to prognosis assessment in invasive transitional bladder cell carcinomas.

The aim of the present work is to ascertain whether additional information to grading and staging can be obtained for the prognosis of invasive bladder tumours (T2, T3, T4) by means of two computer-assisted methodologies. The first methodology relates to the digital image analysis of Feulgen-stained nuclei and the second to a supervised learning algorithm named Decision Tree. The digital image analysis of Feulgen-stained nuclei generated 11 variables for nuclear DNA content and 15 for quantitatively describing chromatin pattern. These 26 variables were submitted to a Decision Tree technique which produces multi-attribute logical classification rules by selecting informative variables and determining discriminatory values for each of them. A series of 41 patients for which the majority of the T2 bladder tumours (68%) were associated with a 'good' prognosis (remission) while the majority of the T3-T4 ones (77%) were associated with a 'bad' one (clinical progression or death) were submitted to the proposed approach. The results show that the decision tree was able to characterise the tumours associated with a 'bad' prognosis in the T2 sub-group (32%) and the tumours associated with a 'good' prognosis in the T3-T4 one (23%), by using only few image-generated variables (added to the clinical stage).