An AI model developed by Alibaba’s DAMO Academy, known as PANDA, has received Breakthrough Device Designation from the FDA for its ability to detect early-stage pancreatic cancer on routine CT scans, becoming the first AI model to earn the designation for this purpose and one of the few Chinese-developed AI diagnostic tools to reach this stage in the U.S. regulatory system. Pancreatic cancer is notoriously difficult to catch early, and PANDA’s developers say the model can spot lesions invisible to the human eye well before symptoms typically prompt a diagnosis.
Why Pancreatic Cancer Is So Hard to Catch
Pancreatic ductal adenocarcinoma is one of the deadliest cancers precisely because it tends to produce no symptoms until it has already spread, by which point curative surgery is often no longer an option. Most cases today are found incidentally or only after a patient develops symptoms serious enough to prompt imaging, long after the disease could have been caught and treated more effectively. That grim survival curve has made early detection one of the most sought-after and elusive goals in oncology, and it is the specific gap DAMO Academy built PANDA to address.
How PANDA Performs
According to reported validation data, PANDA achieved a sensitivity of 92.9% and specificity of 99.9% for detecting pancreatic ductal adenocarcinoma lesions, and the model reportedly outperformed the average performance of radiologists by 34.1 percentage points on sensitivity and 6.3 percentage points on specificity in the same comparison. The tool is designed for opportunistic screening, meaning it can analyze CT scans a patient is already getting for an unrelated reason, looking for subtle textural changes in pancreatic tissue that a radiologist scanning for a different indication might not be specifically looking for.
Real-World Deployment in China
PANDA has already been deployed across hospitals in China, where DAMO Academy says more than 70,000 patients have been screened using the tool. At the Affiliated People’s Hospital of Ningbo University, the model identified two early-stage pancreatic cancer cases among roughly 40,000 individuals screened, cases the company says had been missed by traditional screening approaches. That real-world deployment data, gathered outside a controlled trial, gives DAMO Academy an unusually large evidence base to bring to U.S. regulators compared with many AI diagnostic tools that seek breakthrough status based primarily on retrospective validation studies.
What Breakthrough Status Does and Doesn’t Mean
The FDA’s Breakthrough Device Program is designed to speed up review timelines for technologies addressing life-threatening or irreversibly debilitating conditions, giving companies more frequent and interactive engagement with agency reviewers throughout development. It is not, however, an approval or clearance, and PANDA still needs to complete the formal review process and demonstrate its performance holds up specifically in a U.S. patient population and on U.S. CT scanner hardware and protocols before it could be used clinically in American hospitals.
Cross-Border Questions
Because PANDA was trained and validated primarily on Chinese patient populations and Chinese hospital imaging equipment, some radiologists have raised questions about how directly its reported sensitivity and specificity figures will translate to the more heterogeneous mix of scanner models, imaging protocols, and patient demographics found across U.S. health systems. Geopolitical sensitivities around Chinese-developed AI models being used in critical U.S. healthcare infrastructure add another layer of scrutiny that purely domestic AI diagnostic tools do not typically face, and DAMO Academy will likely need U.S.-specific validation data before hospitals here adopt the tool at scale.
What’s Next
DAMO Academy is expected to use the breakthrough designation to pursue a formal FDA submission, likely paired with additional validation studies run on U.S. imaging data to address concerns about cross-population generalizability. If PANDA’s performance holds up, it would represent a rare case of a screening tool capable of catching one of medicine’s most lethal and hard-to-detect cancers years earlier than current practice typically allows, though the gap between breakthrough designation and a cleared, reimbursed clinical product with proven outcomes remains substantial.
