AI-Powered Endoscopy Helps Patients Avoid Unnecessary Stomach Cancer Surgery

At the gastrointestinal endoscopy unit on the first floor of Gangnam Severance Hospital on the 13th, Professor Kim Ji-hyun of the Department of Gastroenterology was carefully reviewing footage from a gastric endoscopy performed that morning. As she focused on the rapidly changing video controlled by the handle, a white circle appeared on one area of the stomach lining. The numbers at the bottom of the monitor, which had been fluctuating rapidly, turned yellow when they reached 0.7 to 0.8. She was using "Waycen Endo," an AI-powered diagnostic support solution for gastric and colonoscopy that Professor Kim developed with medical AI company Waycen.

Waycen Endo uses AI to analyze endoscopy footage in real time, detecting abnormal lesions such as elevated or depressed areas in the stomach lining and alerting medical staff. The system can identify adenomas, early-stage gastric cancer, and advanced gastric cancer, and can also predict the depth of cancer invasion. This is possible thanks to an AI engine trained through deep learning on both endoscopy images and pathology data. A recent update added a "positive alarm" feature that provides specific numerical values when the AI determines the stomach lining resembles that of patients prone to developing gastric cancer.

"When the probability of stomach cancer increases, the numbers turn red, sending a stronger warning signal," Professor Kim explained. "This prompts the practitioner to examine the lesion more carefully."

The ultimate goal is for AI to change the behavior and approach of endoscopy practitioners, thereby enhancing diagnostic precision.

Stomach cancer, once the leading cancer in Korea, has declined annually and now ranks fifth. According to the 2023 National Cancer Registry Statistics released earlier this year by the Ministry of Health and Welfare and the Korea Central Cancer Registry, there were 28,943 new stomach cancer patients, accounting for 10.0% of all cancer cases. The spread of serving food individually rather than from shared dishes has lowered Helicobacter pylori infection rates, and the widespread adoption of endoscopy screening has increased treatment of gastrointestinal conditions before they progress to cancer. Nevertheless, Korea still has one of the highest stomach cancer rates in the world.

Survival rates for stomach cancer vary dramatically depending on when it is detected. When cancer is found at the "localized" stage, meaning cancer cells have not spread beyond the stomach, the five-year relative survival rate approaches 98%. However, at the "distant metastasis" stage, when cancer has spread to remote organs, the rate drops to just 7.5%. The problem is that more than 80% of early-stage stomach cancer patients experience no noticeable symptoms. Accordingly, the government supports gastric endoscopy costs every two years for adults aged 40 and over through the national health screening program, regardless of symptoms.

Early-stage stomach cancer can be cured through endoscopic resection rather than open surgery. Endoscopic submucosal dissection (ESD) treats early gastric cancer using a specialized endoscope with a disposable blade, similar to removing colon polyps. Since the stomach does not need to be removed, patients can return to daily life quickly and experience significantly improved quality of life.

The key criterion for determining whether endoscopic resection is possible is "invasion depth"—how deeply cancer cells have penetrated the stomach wall. ESD without stomach removal is typically considered when the cancer is confined to the mucosa, the first layer, and is 2 cm or smaller.

Endoscopic ultrasound, which scans the wall to predict how far cancer cells have spread, has traditionally been used to predict invasion depth. However, this method has limitations, as accuracy can vary depending on the practitioner's skill level and the lesion's location. Veterans like Professor Kim can sometimes predict invasion depth just by looking at the cancer's appearance in endoscopy images, but even this approach often leaves ambiguous cases.

"How can we ensure we don't miss a single patient who doesn't need surgery?" This question drove Professor Kim to partner with Waycen to develop AI-based endoscopy diagnostic support software, at a time when applying AI technology in medical settings was still unfamiliar.

"We started from the concept that if we train many endoscopy images on deep learning neural networks, we could predict invasion depth at a level similar to experienced physicians and provide useful information to medical staff," Professor Kim said. "At tertiary hospitals, many patients stand at the crossroads between surgery and endoscopic resection, and this can help medical staff make the best decisions."

Gangnam Severance Hospital has been operating Waycen Endo on all devices in its gastrointestinal endoscopy unit for about four years. According to clinical research recently presented by Professor Kim's team at the Vietnam Gastroenterology Society conference, AI diagnostic performance for early gastric cancer showed 96% sensitivity, 95% specificity, and 96% overall accuracy—significantly higher than endoscopy specialists at 71% accuracy.

The team also obtained research results showing that the rate of recommending unnecessary surgery to patients who could be adequately treated with endoscopy decreased significantly after AI adoption. This demonstrates that AI can possess clinical judgment capabilities at or above the specialist level, not only in early gastric cancer detection but also in predicting lesion invasion depth.

"Humans are easily swayed by preconceptions like 'small lesions must be shallow,' while AI provides objective information based solely on data," Professor Kim noted.

However, this does not mean AI can completely replace specialists. "AI is a partner that helps make more accurate decisions in clinical settings," she emphasized. "Government institutional support is urgently needed for these innovative technologies to be applied to patients as quickly as possible."