Use of artificial neural networks in the clinical staging of prostate cancer: implications for prostate brachytherapy

Abstract

Purpose: This review describes two studies to evaluate artificial neural networks (ANNs) in prostate cancer staging. In the first study, an ANN was trained to identify prostate cancer patients at low risk of lymph node spread (LNS). The second study evaluated an ANN to predict capsular penetration (CP) in men with clinically localized prostate cancer. An accurate assessment of lymph node status will help identify those brachytherapy patients in whom lymphadenectomy can be avoided. The accurate prediction of CP can help determine the appropriateness of brachytherapy as a treatment option.

Materials and methods: An ANN to predict LNS was trained and tested using a database from one institution (n = 4,133) and validated using two databases (n = 330 and n = 227) from different institutions. The clinical variables used were clinical stage (cTNM), Gleason sum, and prostate-specific antigen concentration (PSA). The ANN to predict CP was trained and validated with data from a single institution (n = 409). The variables used were age, race, PSA, PSA velocity, Gleason sum, and cTNM.

Results: The LNS ANN was able classify 76%, 75%, and 30% of the patients in each database as being at low risk of LNS with 98% accuracy. The CP ANN correctly identified CP in 25 (84%) of patients and produced 5 (16%) false-negative predictions.

Conclusions: These preliminary results suggest that ANNs can be useful in staging prostate cancer. If sufficiently accurate ANNs can be developed and tested, they have the potential to increase the accuracy of clinical staging and thus improve treatment decisions.