Hierarchic classification of multiple types of urothelial cells by computer.

A two-level hierarchic classification of 915 urothelial cell images from the urinary sediment was performed by computer using the TICAS programs. The purpose of the analysis was to determine whether adequate discrimination could be obtained among several classes of cells, such as degenerated (DG), multinucleated (MN) and mononucleated well-preserved (WP) ones. The first level in the classification hierarchy showed that the three classes of cells could be identified by computer and that the identification of the group of WP cells was particularly satisfactory, the analysis of WP cells, using training and object sets, documented once again that the identification of diagnostically significant subgroups could be achieved with a very small misclassification error, which was less than 1% for benign (NEG) and malignant (POS) cells. In view of the prior successful application of the classification of images of WP cells to establish patient profiles, the results of the hierarchic classification support the concept of automated analysis of cells in the urinary sediment by computer.