A research team at Asan Medical Center said it has discovered that artificial intelligence can discern whether a colon polyp is benign or malignant, as accurately as “endoscope specialists with five years’ experience” do.

The average AI diagnostic accuracy was 81.8 percent, which is almost similar to that of an endoscopy specialist, 84.8 percent, it said.

In determining whether a colon polyp is benign or malignant, endoscopy specialists first observe the surface and blood vessels of the polyp. If they find it difficult to judge with endoscopic images, they conduct a biopsy, which inevitably causes bleeding or unnecessary polypectomy.

Professor Byeon Jeong-sik (left) of the Department of Gastroenterology and Professor Kim Nam-gook of the Department of Convergence Medicine at Asan Medical Center

The AMG medical team, led by Professors Byeon Jeong-sik and Kim Nam-gook, has developed an AI model, which can judge whether it is positive or malicious, reducing unnecessary biopsies and increasing the diagnosis rate of colorectal cancer by supplementing colonoscopy specialists’ knowledge and experiences, the hospital said in a news release.

Early colorectal cancer that has invaded the mucosal or shallow mucosal layer can be treated with endoscopic resection. In contrast, early colorectal cancer that has invaded the deep mucosal layer must be surgically removed. The AI program is expected to help the medical staff make treatment plans by accurately distinguish the invasion depth of early colorectal cancer.

The second test also showed a similar precision rate to that of the first 82.4 percent.

The pathological diagnosis of the polyps must be accurately predicted to determine whether to resect them with endoscopy, or remove with surgery, or leave them without endoscopy resection, it said.

“Artificial intelligence has also shown significant effectiveness in making judgments by endoscopic trainees with insufficient experience,” Professor Byeon said. “We expect the AI application in colonoscopy will improve diagnostic accuracy, reduce unnecessary biopsies, and provide patients with optimal treatment according to the pathological diagnosis of the polyp.”

Copyright © KBR Unauthorized reproduction, redistribution prohibited