Acute Appendicitis Diagnosis Using a Deep Neural Network Based on Parallel Feature Windows and Dynamic Weighting

Abstract

Acute appendicitis is one of the most common causes of acute abdominal pain requiring immediate surgical intervention. The misdiagnosis rate leading to unnecessary appendectomy has been reported to range from 10 to 30 percent. The objective of this study is to propose a novel deep neural network based model to improve the diagnostic accuracy of acute appendicitis. In this study, the clinical and laboratory data of 4500 patients with suspected appendicitis who were admitted to Gorgan Hospital in Golestan(Iran) between 1400 and 1404 were collected. Eight features, including Age, Sex, Calprotectin, Procalcitonin, RI index, Neutrophil, CRP, and Wbc, were considered as the model inputs. The proposed method is named the Dynamic Feature Windows Network (DFWN), which employs a three branch parallel architecture together with a dynamic weighting mechanism. The inflammation window consists of four inflammatory biomarkers. The bacterial infection window contains two features associated with bacterial infection. The demographic window includes the patient's age and sex. The proposed model was evaluated using 5 fold cross validation. On the test dataset containing 900 samples, the proposed method achieved an Accuracy of 98.00%, a Recall of 98.15%, a Precision of 98.51%, and an F1 Score of 0.9833. These results significantly outperformed the standard CNN with an Accuracy of 95.22%, the conventional DNN with an Accuracy of 92.56%, and XGBoost with an Accuracy of 90.89%.  From a clinical perspective, the proposed approach reduced diagnostic errors by up to 80% compared with XGBoost. The DFWN model, through its three branch parallel architecture and dynamic weighting mechanism, provides an accurate and effective tool for the diagnosis of acute appendicitis. It has the potential to reduce unnecessary appendectomies and improve the quality of clinical decision making in healthcare centres.