The Optimization Variables of Input Data of Artificial Neural Networks for Diagnosing Acute Appendicitis

Abstract

The purpose of this study is to suggest an efficient diagnosis system for acute appendicitis using the artificial neural network model with optimized input variables. Acute appendicitis is one of the most common diseases of the abdomen. However, the accuracy of diagnosis is not high even with experienced surgeons due to its complex symptoms. We used the artificial neural networks model to analyze the complex problems. A total of 801 suspected acute appendicitis patients were collected and a multilayer neural network with thirteen input variables, and two hidden layers with thirty neurons were used to diagnosis acute appendicitis. The mean-square error (0.0011) was stabilized after seven input variables. The nine to thirteen input variables had a high and equal performance (98.81%, 100%, 98.39%, 100%, 99.31%, and 0.995 for specificity, sensitivity, positive predictive value, negative predictive value, accuracy and AUC, respectively). We had optimized the input variables and the performance is significantly higher than the published diagnosis method such as the Alvarado clinical scoring system. We believe that the developed model regarding the multilayer neural network would be a useful method to rapidly and correctly diagnosis acute appendicitis for clinical surgeons.