EN
EN
Latest Research in Nature Medicine: CT Achieves Large-scale Pancreatic Cancer Screening for the First Time Based on DAMO Academy's Medical AI
On November 21, the latest research in *Nature Medicine*, a top international medical journal, showed that through "non-contrast CT + AI", humanity has for the first time obtained a method for large-scale early screening of pancreatic cancer. Alibaba DAMO Academy collaborated with more than a dozen top medical institutions worldwide to apply AI to the screening of pancreatic cancer in asymptomatic populations in physical examination centers and hospitals. Using only the simplest non-contrast CT, 31 clinically missed lesions were detected among more than 20,000 consecutive patients in the real world. Two patients with early-stage pancreatic cancer have already undergone successful surgeries for cure. *Nature Medicine* published a special commentary on this: "Cancer screening based on medical imaging AI is about to enter its golden age."
Pancreatic cancer, known as the "king of cancers", has an average five-year survival rate of less than 10%. It is the malignant tumor with the lowest survival rate in China and even globally. 80% of pancreatic cancer cases are detected at an advanced stage. The disease progresses rapidly and is extremely difficult to cure. Currently, there is a lack of effective screening methods in clinical guidelines because of the high risk of missed or misdiagnoses. The non-contrast CT commonly used in physical examinations and hospitals has low image contrast, making it difficult to identify early pancreatic lesions.
In the latest paper "Large-scale pancreatic cancer detection via non-contrast CT and deep learning", based on the medical AI technology of Alibaba DAMO Academy, institutions such as Shanghai Institution of Pancreatic Diseases, The First Affiliated Hospital of Zhejiang University School of Medicine, Shengjing Hospital of China Medical University, Fudan University Shanghai Cancer Center, General University Hospital in Prague (First Faculty of Medicine, Charles University), Xinhua Hospital Affiliated to Shanghai Jiao Tong University School of Medicine, and Johns Hopkins University proposed for the first time to conduct large-scale early screening of pancreatic cancer through "non-contrast CT + AI".
In response to the characteristics of pancreatic cancer, such as its hidden location and lack of obvious manifestations in non-contrast CT images, the research team constructed a unique deep learning framework, which was finally trained into the early pancreatic cancer detection model PANDA. First, a segmentation network (U-Net) was constructed to locate the pancreas. Second, a multi-task network (CNN) was used to detect abnormalities. Third, a dual-path Transformer was employed to classify and identify the types of pancreatic lesions. In short, this technology uses AI to amplify and identify the subtle lesion features in non-contrast CT images that are difficult to recognize by the naked eye, achieving efficient and safe early detection of pancreatic cancer and overcoming the problem of high false positives in previous screening methods.
Dr. Kai Cao, a co-first author of the paper and from Shanghai Institution of Pancreatic Diseases, introduced that this study constructed the largest CT training set for pancreatic tumors to date (including 3208 real patients). Finally, through multi-center verification in more than a dozen hospitals worldwide, it achieved a sensitivity of 92.9% (accuracy in determining the presence of pancreatic lesions) and a specificity of 99.9% (accuracy in determining the absence of disease). In a retrospective trial of more than 20,000 real cases, 31 clinically missed lesions were detected, and two patients with early-stage pancreatic cancer have already undergone successful surgeries for cure.
Up to now, this technology has been used more than 500,000 times in hospital and physical examination scenarios, with only one false positive in every 1,000 uses. In the future, multi-center prospective clinical verification will continue to be carried out in the hope of changing the pessimistic view that "screening for pancreatic tumors is not recommended".
Professor Yajia Gu, Director of the Department of Radiodiagnosis at Fudan University Shanghai Cancer Center, said that this paper proposed a potential method for large-scale pancreatic cancer screening. While improving the detection rate, it does not impose additional radiation or economic burdens on patients. "Imagine that we can detect whether we have pancreatic cancer by taking the simplest non-contrast CT during a physical examination. This will help many pancreatic cancer patients and reduce the occurrence of tragedies."
Dr. Le Lu, leader of the medical AI team at DAMO Academy and an IEEE Fellow, said that this research is an important milestone, clinically confirming the reliability of the "non-contrast CT + AI" cancer screening technology path. The medical AI team at DAMO Academy is collaborating with many top medical institutions worldwide to explore new methods for low-cost and efficient multi-cancer screening using AI technology, hoping that people can detect multiple early-stage cancers through a single non-contrast CT scan. Currently, this work has achieved phased progress in seven common cancers, including pancreatic cancer, esophageal cancer, lung cancer, breast cancer, liver cancer, gastric cancer, and colorectal cancer. The research results have been published in medical journals such as *Nature Medicine* and *Nature Communications*, as well as top AI conferences such as CVPR, MICCAI, and IPMI.
It is reported that the medical AI team at DAMO Academy has long been committed to the research on the integration of AI and medical imaging, focusing on three major directions: precise cancer diagnosis and treatment, precise diagnosis and treatment of chronic diseases, and pre-screening of neurodegenerative diseases. They have developed a full-process cancer diagnosis and treatment technology including large-scale screening, precise diagnosis, prognostic treatment, and response assessment. During the initial stage of the COVID-19 pandemic, the team developed an AI-assisted diagnostic system for COVID-19 using CT images, and was rated as an advanced collective in national science and technology anti-pandemic efforts by the Ministry of Science and Technology.
Appendix: Link to the *Nature Medicine* article:
