A new artificial intelligence tool named PANDA is making significant strides in the early detection of pancreatic cancer in China. Developed by researchers affiliated with the Chinese tech giant Alibaba, PANDA has already helped identify several cases of this lethal disease, emphasizing the potential of AI in revolutionizing medical diagnostics.
In a striking instance, Qiu Sijun, a 57-year-old retired bricklayer from eastern China, received life-changing news shortly after a routine diabetes checkup. Dr. Zhu Kelei, head of the pancreatic department at the Affiliated People’s Hospital of Ningbo University, informed him that a follow-up was necessary. Although Mr. Qiu feared the worst, the early detection of his pancreatic cancer allowed for successful removal of the tumor.
Pancreatic cancer remains one of the deadliest forms of cancer, with a dismal five-year survival rate hovering around 10%. This low rate is largely due to the challenges associated with early diagnosis; symptoms typically appear only after the disease has advanced significantly. Traditional diagnostic methods, such as contrast CT scans, expose patients to high levels of radiation, making widespread screening impractical. On the other hand, non-contrast CT scans, which use significantly less radiation, often produce less detailed images, complicating the identification of abnormalities.
PANDA, which stands for “pancreatic cancer detection with artificial intelligence,” was designed to analyze these non-contrast CT scans. Since its introduction in November 2024, the tool has examined over 180,000 scans, uncovering approximately two dozen cases of pancreatic cancer, including 14 in early stages. The tool specifically detected 20 cases of intraductal adenocarcinoma, the most prevalent and aggressive form of pancreatic cancer.
Dr. Zhu revealed that many patients identified by PANDA initially sought medical attention for unrelated complaints, such as bloating or nausea. Their CT scans had not raised alarms until flagged by the AI tool, demonstrating the potential lifesaving capacity of this technology. “I think you can 100% say AI saved their lives,” Dr. Zhu stated.
In April 2024, PANDA received “breakthrough device” status from the U.S. Food and Drug Administration, which will expedite its review and facilitate market access. The tool is also undergoing various clinical trials across China, although researchers emphasize the need for additional real-world data to assess the balance between the benefits of early detection and the risks of false positives.
Experts outside of China have expressed skepticism regarding the efficacy of non-contrast CT scans compared to other imaging methods. Ling Zhang, a senior algorithm engineer at Damo Academy, which developed PANDA, acknowledged initial concerns about the tool’s capabilities. To address these, engineers collaborated with a radiologist to annotate the contrast CTs of over 2,000 known pancreatic cancer cases. This process allowed the AI to learn how to identify lesions even in the less detailed non-contrast images.
The results were promising; in a study published in Nature Medicine in 2023, PANDA accurately detected 93% of pancreatic lesions when tested with over 20,000 non-contrast CT scans. “The effectiveness actually surprised us,” Zhang noted.
At the Ningbo hospital, PANDA analyzes scans already ordered for other medical reasons, eliminating extra costs for patients and the hospital. In China, many individuals undergo non-contrast CTs annually, with costs around $25 prior to insurance.
While Dr. Zhu and his team review scans flagged as high-risk, the model cannot yet replace a trained pancreatic specialist. It sometimes misidentifies cases of pancreatitis, and it cannot differentiate whether a tumor originated in the pancreas or metastasized from another organ. Out of approximately 1,400 alerts generated by PANDA, only about 300 warranted follow-up testing.
Dr. Ajit Goenka, a radiologist at the Mayo Clinic, highlighted the importance of minimizing false alarms. He noted that many individuals in Ningbo might have experienced unnecessary anxiety and invasive follow-up tests after receiving alerts, only to find they were healthy.
Dr. Diane Simeone, a pancreatic surgeon at the University of California San Diego, suggested that PANDA may serve as a valuable tool for less experienced clinicians. She pointed out that some tumors flagged by the AI should have been identifiable by well-trained radiologists unaided.
Despite the tool’s success, the Affiliated People’s Hospital of Ningbo University faces challenges in managing follow-ups due to limited staff and aging hardware that struggles with the data demands of PANDA. Additionally, the prevalence of medical corruption in China has led to public mistrust in healthcare providers, which could deter patients from seeking necessary follow-up care.
Nevertheless, Mr. Qiu’s experience highlights the potential of AI in enhancing patient outcomes. Following his tumor removal, he reported feeling healthy and expressed gratitude for the early detection, stating, “He said I was very lucky. So there was nothing else for me to say. I could only be relieved.”
The ongoing development and implementation of AI tools like PANDA signal a transformative shift in how healthcare challenges such as cancer detection may be approached in the future.
