The Use of Multilayer Perceptron Artificial Neural Network in Diagnosis of Acute Appendicitis

Acute appendicitis is the most common cause of admittance of patients with abdominal pain to hospital and appendectomy is the most commonly performed emergency surgery. Despite significant advances in the field of diagnosis, a significant number of negative appendectomies are reported. In this study, the design and evaluation of artificial neural networks to help diagnose acute appendicitis was investigated.
In this descriptive study, variables affecting the diagnosis were identified through literature review. Then, these variables were categorized in the form of a checklist, and evaluated and prioritized by general surgery specialists. The sample size was determined as 181 cases to train and evaluate the performance of neural networks. The database was created using records of patients who had undergone appendectomy during 2015 in Modarres Hospital, Tehran, Iran. Then, different architectures of artificial multilayer perceptron (MLP) neural network were implemented and compared in MATLAB environment to determine the optimal diagnostic performance. Parameters such as specificity, sensitivity, and accuracy were used for network assessment.
Comparison of the optimal output of the MLP with pathological results showed that the sensitivity, specificity, and accuracy of the diagnosis network were 68.8%, 82%, and 78.5%, respectively. Based on the existing standards and the general surgeons’ opinions, the MLP network improved diagnostic accuracy for acute appendicitis.
The designed MLP can model the performance of an expert with acceptable accuracy. The use of this MLP in clinical decision support systems can be useful in the reduction of negative references to medical centers, timely diagnosis, prevention of negative appendectomy, reduction of the duration of hospitalization, and reduction of medical expenses.