Acute appendicitis hybrid decision support system

Acute appendicitis is an inflammation that develops due to occlusion of appendix lumen. It is treated by surgical removal of the inflamed appendix. It can be difficult to diagnose appendicitis. Because its symptoms can also be seen in many other diseases. It is especially difficult to diagnose appendicitis since its location is variable. Unfortunately, some patients with acute appendicitis may be exposed to peritonitis or other complications during their diagnosis process. These patients have higher mortality rates than patients who are diagnosed timely. In this respect, early diagnosis of acute appendicitis is of great importance to the patient. There are various methods for the diagnosis of appendicitis. One such method is machine learning. Machine learning is used in many areas for solving problems that contribute to decision making and prediction by taking advantage of machine learning algorithms, artificial intelligence, mathematics and statistics. Among them, healthcare field is an important sector where machine learning exercise is thought to be beneficial to both physicians and patients when successful results are obtained. In this thesis, it is aimed to develop a decision support system which can help the diagnosis of acute appendicitis. In the development of the system, statistical methods, logistic regression analysis and one of machine learning method, support vector machine (SVM) are used together. The data used in the study includes 220 patients who were admitted to the Emergency Medical Service of KTU Medical Faculty Farabi Hospital with abdominal pain complaint between 2010 and 2016 years and who were or were not diagnosed with appendicitis. This patient data was obtained by examining biochemical laboratory test results and epicrisis reports in a specialist physician consultancy. Logistic regression analysis is used to select qualified data on the parameters of Gender, Age, Leukocyte, thrombocyte, neutrophil, lymphocyte, PDW (Platelet Dispersion Width), MPV (Mean Thrombocyte Volume), CRP (C-reactive protein), glucose, BUN (Blood Urea Nitrogen), creatinine, total protein, albumin, total bilirubin, direct bilirubin, AST (Aspartate aminotransferase), ALT (Alanine aminotransferase), and amylase. Subsequently, data on the qualified parameters were randomly divided into two groups, the train and the test sets. A hybrid decision support system was developed through SVM method, using 132 patient data from the train set. On the 88 test data, 87% sensitivity, 79% specificity and 82% accuracy values were obtained for the SVM model.