Photo/Illutration MicroRNAs in blood is measured with the next-generation sequencer. (Provided by Arkray Inc.)

KYOTO--Researchers here say they developed a highly accurate method to detect early signs of pancreatic cancer using artificial intelligence (AI).

This form of cancer invariably proves fatal because it is notoriously difficult to detect in the initial stages, and there is no effective treatment once it has taken hold.

A team from Kyoto University and medical equipment maker Arkray Inc. in Kyoto developed a tumor identification model via automated machine learning.

“It offers a novel way to find pancreas cancer earlier without causing much damage to the body,” said team member Akihisa Fukuda, an associate professor of gastroenterology at Kyoto University.

The next step for the researchers is to verify the effectiveness of the method on a more widespread scale so they can seek government approval to put the innovative test technique into clinical use.

Each year, around 44,000 people in Japan are diagnosed with pancreas cancer, according to 2020 data from the National Cancer Center Japan. Pancreatic carcinoma ranks fourth among men and third among women on the list of all cancer fatalities by organ.

The survival rate for pancreatic cancer is significantly higher when the tumor is discovered early. But about half of all patients are found at stage 4, the most advanced level, because the tumor often remains asymptomatic in its initial phases.

No effective diagnostic technique has been developed yet to pinpoint early-stage pancreatic cancer.

The team tested microRNAs, very short fragments of ribonucleic acid (RNA), which are responsible for regulating genetic functions. Of more than 2,500 microRNAs in blood, 100 especially common types were selected to combine with a conventional tumor marker to create a diagnostic model.

The data of the microRNAs and the cancer marker were collected from 93 healthy individuals as well as 92 pancreas cancer patients at 17 medical centers, including Kyoto University Hospital.

The collected data was “fed” into an AI system for automated machine learning to perfect a cancer detection model. The finished diagnostic model was used on an additional 240 patients to gauge its effectiveness.

The detection rate under the newly developed method went up to an impressive 83 percent in Stage 1 cases, while only 29 percent of patients with pancreatic cancer tested positive with the tumor marker alone at this low level of progression.

Getting a positive reading for pancreatic cancer with the AI method went from zero to 50 percent even in the very early stage of the disease when spotting the cancer is generally difficult.

The findings indicate that the team’s method can prove particularly helpful in discovering less advanced pancreatic cancer.

“The survival rate could improve dramatically through surgery in stages zero and 1, whereas patients with more advanced stages often struggle to recover,” said Fukuda.

In stage 2 and above, the success rate for the new technique ranged from 93 percent to 97 percent. The overall result clocked in at 90 percent.

Arkray said it is aiming to pitch the prototype for research to the market within one to two years.

The team’s research outcomes were published in the specialized tumor magazine British Journal of Cancer at (https://www.nature.com/articles/s41416-024-02794-5).

Related News