Computer identification of multinucleated urothelial cells.

Computer analysis of multinucleated urothelial cells from the urinary sediment is reported. The cell images were analyzed by two methods: the first based on nucleus-finding algorithms (determination of nuclear boundaries) and the second based on algorithms derived from the histograms of optical density. Both approaches resulted in good discrimination between benign and malignant cells, although the misclassification rate rose slightly from +/- 7% by the first approach to +/- 10% by the second approach. Algorithms identifying multinucleated cells are now a part of the TICAS file as an important step in automated analysis of cells in the urinary sediment.