Prediction of Upper Urinary Tract Calculi Using an Artificial Neural Network
Monthira Tanthanuch MD*, Sawit Tanthanuch BEng, MEng**
Affiliation : * Department of Surgery, Faculty of Medicine, Prince of Songkla University ** Department of Electrical Engineering, Faculty of Engineering, Prince of Songkla University
Objectives : To evaluate the possibility of using an artificial neural network (ANN) in upper urinary tract
calculi prediction.
Materials and Methods : Data of 168 upper urinary tract calculi patients treated in the Division of Urology,
Department of Surgery, Songklanagarind Hospital from January 1997 to December 2000 were reviewed and
classified into 6 catagories and 20 characteristics. 100 items were used in training and 68 in testing for an
ANN designed with 3 layers: 20 nodes for an input layer, 5 nodes for a hidden layer and a node for the output
layer.
Results : Output data between 0-0.38 indicate free of calculi, 0.65-1 indicate prone to have calculi, 0.38-
0.65 indicate probable calculi and further need investigation.
Conclusion : An ANN with error back-propagation training can be used in diagnosing the presence of upper
urinary tract calculi. The accuracy of prediction depends on a previous history of calculi, nephrocalcinosis,
24 hour urine assay for citrate and urine culture.
Keywords : Urinary tract calculi, Neural network
