Acute appendicitis is a common cause of acute abdominal pain in the emergency department, accounting for more than 10% of acute abdominal pain consultations. Due to symptoms that are similar to other abdominal conditions, traditional diagnoses relying on a doctor's experience often carry the risk of misdiagnosis or delayed treatment.
To address this issue, Dr. Kuei-Hong Kuo’s team at Asia University Hospital developed an artificial intelligence-assisted diagnostic system for acute appendicitis using abdominal CT images. This system leverages advanced AI technology and employs a 3D nnUNet deep learning algorithm to accurately identify acute appendicitis, significantly improving diagnostic accuracy and efficiency.
The 3D nnUNet Architecture Enhances Accurate Localization, Upgrading Appendicitis Diagnosis
The core technology of this AI system is the 3D nnUNet architecture, a deep learning model specifically designed for medical image segmentation. It offers exceptional flexibility and adaptability, capable of processing images with varying resolutions. nnUNet has demonstrated outstanding performance in multiple international medical image segmentation challenges and has successfully been applied across various fields, proving its excellence in medical imaging.
Compared to traditional imaging diagnostic systems, this system's advantages lie not only in its algorithmic structure but also in its data set. The 1,800 detailed annotated cases of acute appendicitis provided by Asia University Hospital serve as a crucial foundation for training the model. In addition to acute appendicitis cases, the system also utilized 3,000 image datasets related to abdominal pain, including acute diverticulitis and intestinal infections, which share highly similar clinical manifestations and imaging characteristics with appendicitis. Through such an “actively filtered” dataset, the system can provide greater accuracy when identifying similar diseases, reducing the risk of misdiagnosis or missed diagnosis.
Additionally, the system’s acute appendicitis diagnostic algorithm has been specially optimized to enhance its sensitivity in detecting swelling while accurately identifying the relative position of the appendix in relation to other abdominal organs. This is particularly important in clinical practice, as the location of the appendicitis lesion can sometimes be affected by anatomical variations in patients, complicating diagnosis. However, the system successfully handled such situations in testing, demonstrating its strong ability to manage complex anatomical features. This enables the AI system not only to quickly locate lesions but also to significantly reduce the risk of misdiagnosis by physicians and improve diagnostic accuracy.
Dr. Kuei-Hong Kuo Shares Technological Achievements and Future Plans
Dr. Kuei-Hong Kuo noted that the development of this AI system represents not only a technological breakthrough but also a successful practical application. He emphasized that the system’s greatest feature lies in its ability to provide accurate and rapid diagnostic results for physicians, reducing subjective factors and delays in the diagnostic process, while offering substantial support for high-pressure environments such as the emergency department. He added that with the successful implementation of the system at Asia University Hospital, the next step will be to conduct larger-scale, multi-center clinical trials, with plans to complete TFDA certification by mid-2025, further advancing commercialization. Looking ahead, the system is not only expected to serve the Taiwan market but also plans to expand into other regions of Asia, as well as the United States and Europe. Particularly in remote areas with relatively scarce medical resources, the application of this system is expected to significantly enhance the diagnostic efficiency of acute appendicitis and improve patient treatment outcomes.
Resource: AI辨識影像精準定位急性闌尾炎 提升診斷準確性
