COMPARING NEURAL NETWORKS IN THE DISCRIMINATION OF BENIGN FROM MALIGNANT LOWER URINARY-TRACT LESIONS

Objective To compare the performance of two different neural networks (NNs) in the discrimination of benign and malignant lower urinary trac t lesions. Materials and methods A group of patients was evaluated, co mprising 50 cases of lithiasis, 61 of inflammation, 99 of benign prost atic hyperplasia (BPH), five of in situ carcinoma, 71 of grade I trans itional cell carcinoma of the bladder (TCCB), and 184 of grade II and grade III TCCB. Images of routinely processed voided urine smears were stained using the Giemsa technique and analysed using an image-analys is system, providing a dataset of 45 452 cells. Two NN models of the b ack propagation (BP) and learning vector quantizer (LVQ) type were use d to discriminate benign from malignant cells and lesions, based on mo rphometric and textural features. The data from 13 636 randomly select ed cells (30% of the total data) were used as a training set and data from the remaining 31816 cells comprised the test set. Similarly, in a n attempt to discriminate patients, 30% of the cases, selected randoml y, were used to train a BP and an LVQ NN, with the remaining 329 cases used for the test set. The data used for training and testing were th e same for the two kinds of classifiers. Results The two NNs gave simi lar results, with an overall accuracy of discrimination of approximate to 90.5% at the cellular level and of approximate to 97% for individu al patients. There were no statistically significant differences betwe en the two NNs at the cellular or patient level. Conclusions The use o f NNs and image morphometry could increase the diagnostic accuracy of voided urine cytology; despite the different nature of the two classif iers, the results obtained were very similar.