Discriminant analysis of exfoliated cells in bladder urothelium cancer

Abstract

Objective To explore the method of computer discriminant on exfoliated cells in bladder urothelium cancer by Pap stain. Methods The exfoliated cells in 107 urine smears included 386 uroepi-thelium normal exfoliated cells (UNC) , 439 urothelium dysplastic exfoliated cells (UDC) and 500 blad-der urothelial cancer exfoliated cells (UCC). The cells were randomly divided into training group (n = 1077) and identifying group (n = 248) , and the chromatic and geometric shape parameters of cytoplasm and nuclear were tested. The stepwise discriminant analysis was used in cells of the training group to establish a discriminant function and analyze the rate of back substitution discriminant. The function was evaluated by cells of identifying group, and the coincidence rate was analyzed in 107 specimens. Results The back discriminant coincidence rate of cells in training group was 80.8%. The coincidence rate of identifying group and 107 specimens were 80.2% and 92.5% respectively. The discriminant effect was significantly better than function based on chromatics and geometric shape parameters individually (P < 0.05). Conclusions The function combined with chromatics and geometric shape parameters has good discriminant performance in bladder urothelium cancer.