Development of an AI model for automatic diagnosis of appendicitis with CT image reading… “Diagnostic accuracy 89.4%”

An artificial intelligence (AI) model that automatically diagnoses appendicitis by analyzing computed tomography (CT) images has been developed by domestic medical staff. The accuracy of the AI ​​model in diagnosing pendicitis was found to be close to 90%.

Acute pendicitis has variable clinical features, and is sometimes misdiagnosed as another digestive disease because an abnormal appendix is ​​not detected even by a CT scan. If this AI model is used in practice, it is expected that there will be less misdiagnosis of appendicitis and faster patient care will be possible. It is also expected to help operate emergency room personnel more efficiently.

Hallym University Sacred Heart Hospital Surgery Research Team and Hallym University Medical Center Medical Artificial Intelligence Center recently developed an AI model that automatically detects appendicitis by observing CT images in real time.

Appendicitis, commonly known as appendicitis, refers to inflammation of the appendix, which is the end of the appendix. Symptoms include nausea, vomiting, nausea, etc., followed by gradually increasing pain intensity in the epigastric region and upper abdomen. Over time, pain in the upper abdomen goes around the navel and changes to pain in the lower right abdomen, which is the location of the appendix.

Acute appendicitis is a very frequent disease that ranks fifth in surgical statistics, and it is also a disease that can cause misdiagnosis.

Due to the nature of the disease, patients who visit the hospital with suspected symptoms of acute appendicitis often visit the hospital through the emergency room at night or on weekends. In this case, an accurate reading by an abdominal radiologist may be limited. In addition, acute appendicitis has a variety of clinical features, and there are cases where an abnormal appendix is ​​not detected even by CT imaging, so it is misdiagnosed as another digestive disease.

The problem is that if the diagnosis of appendicitis is delayed, perforation may occur, and if inflammation in the right lower abdomen of appendicitis develops into peritonitis or intrapelvic abscess, surgical treatment beyond appendectomy may be derived from that. In addition, postoperative complications increase.

This AI model developed by the research team at Hallym University’s Sacred Heart Hospital observes CT images in real time to filter out diseases that are clinically similar to pendicitis, such as colitis, terminal ileitis, and ascending colonic diverticulitis, and correctly diagnose appendicitis only.

The research team analyzed data from 4701 patients who underwent CT scans for the treatment of appendicitis at Hallym University Medical Center between 2013 and 2020 and data from 4452 patients who visited the emergency room and underwent CT scans for pain in the abdomen between 2019 and May this year. .

Subsequently, the data of 1839 patients with appendicitis and 1782 patients diagnosed as not appendicitis were filtered out and trained in a model using a ‘3D Convolutional Neural Network (CNN)’.

The appendicitis diagnosis accuracy of the AI ​​model that completed training was 89.4%.

The ‘Area Under the Curve’ (AUC) score used to evaluate the performance of the AI ​​model was 0.890, showing excellent results that can be applied to actual clinical practice.

“This AI is significant because it recognizes three-dimensional CT images in three dimensions, unlike existing models,” said Beomju Cho, head of the Center for Medical Artificial Intelligence.

Professor Son Il-tae said, “We are reviewing different methods to increase the sensitivity, area under the curve score, and F1 score of this AI model.” Mr. said

This AI model was presented at the Korean Society of International Surgery Fall Conference and the Korean Society of Surgeons Fall Conference and won the ‘Best Principal Investigator’ award. Reporter Jang Jong-ho [email protected]

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.