Please view the main text area of the page by skipping the main menu.

Medical AI holds out hope for speedy, accurate diagnosis of stomach cancer precursor


Artificial intelligence (AI) appears on the cusp of taking a major role in the doctor's office, with the development of systems including AI-based endoscopic imaging diagnosis that perform at least as well as their medically expert human counterparts.

In addition to reducing doctors' workloads and helping prevent diseases being overlooked, the technology is expected to have applications in pharmaceutical development and other medical fields. However, some questions need to be answered before AI can be adopted across the medical community, such as where responsibility lies if an AI system makes a mistake.

One paper published in October in an industry journal shocked the medical community. It stated that an AI-based tool for endoscopic imaging diagnoses -- developed by the Tada Tomohiro Institute of Gastroenterology and the Osaka International Cancer Center, among others -- had achieved the same proficiency as expert physicians in identifying Helicobacter pylori bacterial infections, a stomach cancer precursor.

The AI system had analyzed at least 10,000 endoscopic images from about 400 patients in just over three minutes, and determined whether each person had or did not have H. pylori with nearly 90 percent accuracy. Meanwhile, 23 doctors expert in endoscopic diagnoses took an average of four hours to go through the same images, and only three made more successful diagnoses than the AI.

"The accuracy of the AI's diagnoses is amazing," one 34-year-old physician specializing in endoscopies said. "I'm no match for that kind of speed."

The vast majority of stomach cancer cases in Japan are caused by H. pylori infections. In 2016, the Ministry of Health, Labor and Welfare revised its stomach cancer diagnosis guidelines to add endoscopic examinations to X-rays as a diagnostic tool. Endoscopies allow doctors to confirm even small medical problems and perform biopsies, while also negating the need to bombard patients with X-ray radiation. The number of people opting for endoscopies is expected to rise, leading to a virtual flood of diagnostic images.

However, the only visible sign of a H. pylori infection is a slight reddening of the gastric mucosa, the stomach's mucous membrane, making it difficult to identify. Correct diagnosis of the condition relies heavily on the competence of the doctor.

There are only about 1.3 endoscopic examination experts per 10,000 people in Japan, meaning measures are needed to both reduce the experts' workload and prevent medical problems from being overlooked.

About 50,000 people underwent endoscopies last year for stomach cancer screening within the jurisdiction of the Urawa Medical Association in Saitama Prefecture, where Tada Tomohiro Institute of Gastroenterology head Tomohiro Tada is based. About 40 diagnostic images were taken per patient, meaning a total of some 2 million images had to be checked. The diagnostic AI is expected to be able to sift through all that in about half a day, while maintaining a high level of accuracy.

"If the AI and medical experts work together to check difficult-to-identify conditions, then we can reduce the chances of a missed diagnosis," Tada told the Mainichi Shimbun. "AI should become indispensable to endoscopic image diagnoses."

Also in The Mainichi

The Mainichi on social media