Predicting Risk of Acute Appendicitis: A Comparison of Artificial Neural Network and Logistic Regression Models

Abstract

Acute appendicitis is considered as one of the most prevalent diseases needing urgent action. Diagnosis of appendicitis is often complicated, and more precision in diagnosis is essential. The aim of this paper was to construct a model to predict acute appendicitis based on pathology reports. The analysis included 181 patients with an early diagnosis of acute appendicitis who had admitted to Shahid Modarres hospital. Two well-known neural network models (Radial Basis Function Network (RBFNs) and Multi-Layer Perceptron (MLP)) and logistic regression model were developed based on 16 attributes related to acute appendicitis diagnosis respectively. Statistical indicators were applied to evaluate the value of the prediction in three models. The predicted sensitivity, specificity, positive predicted value, negative predictive values, and accuracy by using MLP for acute appendicitis were 80%, 97.5%, 92.3%, 93%, and 92.9%, respectively. Maine variables for correct diagnosis of acute appendicitis were leukocytosis, sex and tenderness, and right iliac fossa pain. According to the findings, the MLP model is more likely to predict acute appendicitis than RBFN and logistic regression. Accurate diagnosis of acute appendicitis is considered an essential factor for decreasing mortality rate. MLP based neural network algorithm revealed more sensitivity, specificity, and accuracy in timely diagnosis of acute appendicitis.

Keywords

• Acute appendicitis
• Neural network
• Multi-layer perceptron
• Radial-based function
• Logistic regression