Alibaba Group’s research institute DAMO Academy said it has come up with an early screening method powered by AI technologies to detect pancreatic cancer with high-accuracy, a significant breakthrough that made large-scale pancreatic cancer screening possible.
Its deep learning-based algorithm can detect pancreatic lesions hard to be observed by human eyes in non-contrast CT scan, making it possible to enhance imaging-based screening for pancreatic cancer.
According to a recent journal from Nature Medicine, the model, trained on more than 3,200 image sets, achieved high performance on key diagnostic indicators. It achieved a specificity of 99.9%, suggesting that there’s only one false-positive in every 1,000 tests and a sensitivity of 92.9%, outperforming human radiologists by 34.1% in sensitivity and 6.3% in specificity.
In collaboration with over ten world-leading medical institutions, researchers from DAMO Academy used the AI-based screening method to screen over 20,000 patients and detected 31 cases of pathological changes that were earlier missed by doctors. The model has by far been used over 500,000 times in hospital and medical check-up settings in China.
“Early detection of pancreatic cancer is hard to realize in conventional screening, which results in late detection and poor prognosis. The AI plus non-contrast CT technology hold the promise to be an effective and cost-efficient tool to achieve detection of the pancreatic cancer in the early stages and make large-scale pancreatic cancer screening possible to prevent the loss of lives,” said Le Lu, head of Alibaba's Damo Academy's medical AI team and fellow at IEEE.
The survival rate of pancreatic cancer is low compared to other cancers, partly because it’s often found at later stages when treatment is hard. It is the seventh leading cause of cancer-related deaths worldwide, with an average five-year survival rate of around 5 to 10 percent.
Combined with non-contrast CT imaging, the early screening technology can help doctors with early detection of pancreatic cancer, a challenging disease given its often unspecific symptoms. It can also be applied into large-scale screening efforts, for example, as part of the non-contrast CT offering in routine medical checkups or during visits to emergency departments.
The accuracy metrics of the algorithm are “superior to those of several acknowledged screening methods such as Pap smears for cervical cancer or mammography for breast cancer. This makes it tempting to call for integration of this specific method into large-scale screening effort…AI-based screening is a highly promising approach with potential for clinical impact in the near future,” said Jörg Kleeff & Ulrich Ronellenfitsch from Martin-Luther-University Halle-Wittenberg, University Medical Center Halle (Saale) at Germany. The professors also pointed out further thorough assessment is needed before it is ready to be rolled out into wide- spread practice.
