| Summary: | Aim: improvement of early diagnostics of prostate cancer by developing a technique, which makes possible to predict its clinical significance in outpatient setting before initiation of invasive procedures. Material and Methods. Clinical data of 398 patients who underwent transrectal prostate biopsy in 2012-2014 in SSMU S. R. Mirotvortsev Clinical Hospital, was used to build an artificial neural network, while its output allowed to determine whether the tumour corresponds to Epstein criteria and which D'Amico risk group it belongs to. Internal validation was performed on 80 patients, who underwent prostate biopsy in September 2014 — December 2014. Sensitivity, specificity, positive and negative predictive value of artificial neural network were calculated. Results. Accuracy of predicting adenocarcinoma presence in biopsy specimen was 93,75%; accuracy of predicting whether the cancer meets active surveillance criteria was 90%. Accuracy of predicting T stage (T1c, T2a, T2b, T2c)was 57,1%. Prediction of D'Amico risk group was accurate in 70% of cases; for low-risk cancer accuracy was 81,2%. Conclusion. Artificial neural networks may be responsible for prostate cancer risk stratification and determination of its clinical significance prior to biopsy.
|
|---|
