SIU 2017: Artificial Intelligence (AI) Can More Efficiently Predict Prostate Cancer Compared with PSADT and PSAD

Lisbon, Portugal ( As all Urologists know, patients with fluctuating PSA’s can be very difficult to manage. Often times, there are preconceived notions of risk that influence our decision to biopsy, obtain genomic testing, or intervene. However, as with any other human based decision, inter operator variability is high – the same patient presented to two different urologists may result in different outcomes, as much as we would like to think guidelines should reduce variability.

In this study, the authors attempt to introduce the concept of artificial intelligence and computer learning to help manage patients with elevated PSA. Specifically, they use three different AI tools – Neural network back progression (BP), support vector machine (SVM), and random forest (RF).

To do so, the authors identified 3911 patients who underwent biopsy between 2002 and 2016. For those patients, 2 years of PSA data and biopsy pathology data were obtained and reviewed. 657 patients had 2 years of continuous PSA data and 491 of them had biopsy pathology data. Age, PSA data (min, max, median, variance), prostate volume, and absence/presence of pyuria were entered as input data. The primary outcome was accurate prostate cancer diagnosis (any prostate cancer) – ROC and AUC were then calculated for each method.

Prostate cancer was identified in 250 of the 657 men with continuous PSA data, and in 193 of the 491 men with prostate biopsy. The accurate diagnosis of each method was calculated for both groups. The accuracy was 69% for the 657 men with continuous PSA data and 71-72% for the 491 men with biopsy data as well, for all three methods. 

Interestingly, when compared to PSAV and PSAD (thought cutoffs were not included in the poster), the AUC of each of those methods was superior to PSAV and PSAD. Generally, the AUC was 0.75 to 0.78, but for PSAV and PSAD it was 0.53 and 0.45.

These are very interesting results. They are conducting a prospective analysis of these techniques at this time. 

Presented by: Masakazu Tsutsumi
Affiliation: Hitachi General Hospital, Japan

Written by: Thenappan Chandrasekar, MD, Clinical Fellow, University of Toronto, Twitter: @tchandra_uromd at the 37th Congress of Société Internationale d’Urologie - October 19-22, 2017- Lisbon, Portugal