Comparison between artificial neural networks and urologists' assessment of outcome in bladder cancer - Part I: Progression and recurrence in Ta/T1 tumours

The early accurate determination of course of disease in Ta/T1 bladder cancers is an important issue in patient management and improvement of clinical outcome. For this purpose a comprehensive database of patients with newly diagnosed bladder cancer was retrospectively analyzed by artificial neural networks (ANNs) as follows. First, stage progression in 105 patients with Ta/T1 tumours was analyzed using 7 different factors including clinicopathological and molecular markers of mixed prognostic significance. Eight additional factors were then employed to analyze tumour recurrence within 6 months in 56 patients. The prediction accuracies of the ANNs were subsequently compared to those of 4 expert urologists and proved to be significantly higher in predicting stage progression. An important result of the analysis concerned the T1G3 group of tumours which is non-infiltrative at diagnosis, but has the greatest propensity to progress to muscle-invasive disease. In this group, again, the performance of the ANN exceeded that of the urologists.