内容摘要：By Zhou Wenting (Chinadaily.com.cn) 08:55, November 23, 2023A deep learning approach now makes it po

By Zhou Wenting (Chinadaily.com.cn) 08:55, November 23, 2023

A deep learning approach now makes it possible to detect and classify pancreatic lesions with high accuracy via non-contrast computed tomography, and it "could potentially serve as a new tool for large-scale pancreatic cancer screening", according to a paper published in a leading medical journal on Monday.

The approach, pancreatic cancer detection with artificial intelligence, is based on deep learning technology developed by Alibaba Group's Damo Academy.

The paper was published on the website of the medical journal Nature Medicine.

Researchers from Damo Academy and more than 10 prestigious medical institutions in China, the Czech Republic and the United States used medical AI technology and CT scans to detect 31 cases of pathological changes while screening over 20,000 real-world asymptomatic individuals for pancreatic cancer. Among them, two patients with early-stage pancreatic cancer went on to be cured by surgery.

The average five-year survival rate of patients diagnosed with pancreatic cancer is less than 10 percent, making the cancer a malignant tumor with one of the lowest survival rates both in China and worldwide. Around 80 percent of pancreatic cancer cases are only detected at an advanced and inoperable stage.

Medical experts said that there is a lack of effective screening methods in the current clinical guidelines, as the contrast of CT scan images commonly used in physical examinations is low, which makes it hard to identify early pancreatic pathological changes.

In view of the often hidden location of pancreatic tumors and the lack of obvious representation in CT images, researchers have constructed a deep learning framework and developed it as an early detection model for pancreatic cancer. Among its functions are locating the pancreas, detecting abnormalities, and classifying and identifying the types of pancreatic pathological changes.

"In short, the technology uses AI to magnify and identify the subtle features of pathological changes in non-contrast CT images that are difficult to identify with the naked eye and thus achieves efficient and safe early pancreatic cancer detection. It also overcomes the problem of high false positives as seen in earlier screening methods," said Lyu Le, who is in charge of the medical AI team at Damo Academy.

Cao Kai, co-first author of the paper and a doctor at the Shanghai Institute of Pancreatic Diseases, said that the study was verified by more than 10 hospitals, and showed 92.9 percent sensitivity, the rate of accuracy in determining the presence of pancreatic pathological changes, and 99.9 percent specificity, the rate of accuracy in determining the absence of the disease.

The institutions involved in developing the approach include the Shanghai Institute of Pancreatic Diseases, the First Affiliated Hospital of Zhejiang University School of Medicine, Shengjing Hospital of China Medical University, the First Faculty of Medicine at Charles University in Prague, and Johns Hopkins University in the US. Researchers said that they will continue to conduct multi-center, prospective clinical validation.

"The paper proposed a potential method to screen for pancreatic cancer on a large scale. It may improve the detection rate without putting additional radiation and financial burdens on patients," said Gu Yajia, director of the department of diagnostic radiology at the Fudan University Shanghai Cancer Center.

The medical AI team at Damo Academy said it is also collaborating with multiple top medical institutions around the world to use AI to explore new methods of low-cost and efficient multiple cancer screening, in order to allow individuals to screen for a variety of early-stage cancers through a single non-contrast CT scan.