◎ Reporter Zhang Qiang and correspondent Bai Jin
Pancreatic cancer is known as the "king of cancers" and has the lowest five-year survival rate of all malignancies. However, it is difficult to detect pancreatic cancer with common noncontrast CT. Now, with the assistance of artificial intelligence (AI), non-contrast CT is expected to play a huge role in the early screening of pancreatic cancer on a large scale. Recently, the relevant research results were published in the form of original papers in the international medical journal Nature Medicine, with Cao Kai, attending doctor of the Department of Diagnostic Radiology of the First Affiliated Hospital of Naval Medical University (Shanghai Changhai Hospital), as the co-first author, and Professor Shao Chengwei and Professor Lu Jianping as the co-corresponding authors.
As a gastrointestinal cancer with a very poor prognosis, pancreatic cancer has clinical characteristics such as difficult early diagnosis, low surgical resection rate, and easy recurrence and metastasis after surgery. "The early symptoms of pancreatic cancer are insidious, and there is still a lack of biomarkers or imaging methods with high sensitivity and specificity that can be widely used in large-scale population screening." "This leads to a low early diagnosis rate of pancreatic cancer, and once diagnosed, 80% of pancreatic cancer patients are in the middle and advanced stages. ”
Cancer treatment focuses on early screening and early treatment, and it is very important to improve the early screening rate of pancreatic cancer to improve the prognosis of pancreatic cancer patients. The early detection and treatment of pancreatic cancer has always been the focus of attention in the medical community. Chest noncontrast CT is simple and easy, and is widely used in the screening of pulmonary nodules, and has become an internationally recognized appropriate and high detection rate of early lung cancer screening methods. So, can chest noncontrast CT moderately expand the scope of application for pancreatic cancer screening?
Led by the Shanghai Institute of Pancreatic Diseases, the research team worked with the Ali Damo Academy, the First Affiliated Hospital of Zhejiang University School of Medicine and other institutions to build a unique deep learning framework, which was finally trained as a pancreatic cancer detection model (PANDA). According to Cao Kai, PANDA uses a segmentation network (U-Net) to locate the pancreas, a multi-task network (CNN) to detect lesions, and a dual-channel Transformer module to distinguish pancreatic cancer from other pancreatic lesions. To put it simply, the "three-step method" is to use AI to magnify and identify those subtle pathological features in non-contrast CT images that are difficult to identify with the naked eye.
The CT training set of pancreatic tumors constructed by this model has included 3208 surgical cases, and has been verified by 10 hospitals around the world, with a sensitivity of 92.9% (accuracy in judging the presence of pancreatic tumors) and 99.9% specificity (accuracy in judging the absence of tumors). In a real-world retrospective trial of 20530,31 people, the model identified 2 clinically missed lesions, of which <> patients with early-stage pancreatic cancer had completed surgical treatment.
"The results of this study fully demonstrate that the use of 'non-contrast CT+AI' for large-scale early screening of pancreatic cancer has great potential. This will provide new support for optimizing the diagnosis and treatment guidelines for pancreatic cancer screening, and will also have a positive impact on the diagnosis and treatment process, treatment decisions, and treatment costs of pancreatic cancer in the future. Shao Chengwei said.
"This study clinically confirms the reliability of the 'non-contrast CT+AI' cancer screening technology pathway, and provides an innovative strategy for the treatment of pancreatic cancer." Cao Kai said that with the continuous maturity and promotion of technology, "non-contrast CT+AI" may be included in the physical examination program in the future, which will greatly improve the early screening rate of pancreatic cancer.