Summary
Aim – evaluate the diagnostic capabilities of an artificial neural network to identify neoplastic changes and the potential to optimize diagnostic methods.
Methods. The prognostics of uterine endometrioid adenocarcinoma was carried out by the artificial neural network, logistic regression, criteria-based prognostic models. The prognostic potential was assessed by analyzing 416 cases of surgical samples after total hysterectomies pathology testing in the Pathological anatomy department of Lithuanian University of Health Sciences hospital Kaunas Clinics. Models’ prognostic potential was evaluated with ROC, AUC. Cohen’s kappa coefficient determined each model’s correspondence to pathology report conclusions.
Results. Logistic regression model was more accurate than criteria-based model when detecting endometrioid adenocarcinoma’s cases (AUC=0.842; k=0.55, 95% CI 0.45–0.642; p<0.001). ANN was the most accurate prognostic tool for detection of endometrioid adenocarcinoma cases (AUC=0.908; k=0.594, 95% CI 0.504–0.684; p<0.001).
Conclusions. The accuracy of pathology diagnostics differed among separate meth ods of data analysis. In the future, artificial in telligence could be helpful to identify unique structural changes, possible patterns to help create new diagnostic methods. Artificial intelligence proved useful to optimize pathology diagnostic to yield the best accuracy.
