A team of researchers from Taiwan has developed PanMETAI, an AI-powered platform that analyzes metabolic fingerprints in a simple blood sample to detect pancreatic cancer at its earliest stages—when treatment is most effective—achieving up to 94% diagnostic accuracy.
The challenge of catching pancreatic cancer early
Pancreatic cancer is one of the deadliest forms of cancer worldwide, with only a 13% five-year survival rate. The disease is notoriously difficult to catch early: most patients receive their diagnosis at an advanced stage, when treatment options are limited. Current screening methods, including the widely used blood marker CA19-9, lack the sensitivity and specificity needed for reliable early detection.
Now, a cross-disciplinary research team from National Taiwan University Hospital and Academia Sinica has developed a breakthrough diagnostic tool called PanMETAI. Published in Nature Communications, the study describes how the platform combines artificial intelligence with nuclear magnetic resonance (NMR) metabolomics—a technique that captures the unique chemical fingerprint of hundreds of metabolites in a patient's blood—to identify pancreatic cancer with remarkable accuracy.
How the PanMETAI platform works
Using just 500 microliters of blood serum, the platform extracts over 260,000 metabolic signals and analyzes them through a state-of-the-art tabular foundation model called TabPFN.
By integrating these metabolic profiles with age, the cancer marker CA19-9, and a protein biomarker called Activin A, PanMETAI achieved an area under the curve (AUC) of 0.99 in the Taiwanese cohort—meaning it correctly distinguished cancer patients from high-risk controls in nearly every case.
Crucially, the model was also validated in an independent Lithuanian cohort of 322 participants from Lithuanian University of Health Sciences—a completely different ethnic and geographic background. There, it maintained strong performance with an AUC of 0.93, demonstrating that the tool works reliably across diverse populations—a common hurdle for medical AI systems.
Strong performance in early-stage detection
One of PanMETAI's most significant achievements is its ability to detect early-stage (Stage I/II) pancreatic cancer, a challenge that has long eluded researchers.
The study found that NMR metabolomic data were essential to boosting early-stage detection sensitivity, capturing subtle metabolic shifts—such as decreased HDL cholesterol and glutamine, and elevated lactic acid, glucose, and glutamic acid—that occur before the cancer becomes clinically apparent.
Potential impact on clinical practice
The platform also performs well even with very small training datasets. In testing, it achieved stable accuracy of approximately 90% with as few as 50 training cases, making it a practical solution for hospitals and research centers that may not have access to large patient cohorts.
The research team envisions PanMETAI as a rapid, non-invasive, and cost-effective screening tool that could be deployed in clinical settings to flag high-risk individuals for further evaluation, potentially saving lives through earlier intervention.
Dr. Chun-Mei Hu, Assistant Research Fellow at the Genomics Research Center, Academia Sinica, and co-corresponding author of the study, said, "Our study demonstrates how a triple-synergy between clinical expertise, basic cancer research, and advanced AI can bridge the gap between laboratory discovery and diagnostic application."
"By integrating Professor Yu-Ting Chang's clinical insights with our metabolic molecular findings and Dr. Chao-Ping Hsu's application of machine learning and AI, we developed PanMETAI. This platform moves beyond the traditional 'black box' of AI into a transparent discovery engine, enabling high-precision pancreatic cancer prediction across international cohorts and bringing early detection and timely intervention within reach for patients."
Dr. Chao-Ping (Cherri) Hsu, Distinguished Research Fellow at the Institute of Chemistry, Academia Sinica, noted, "We are witnessing a new era where AI and medicine converge. Our work demonstrates the power of machine learning to navigate complex data, integrating high-density spectra with essential clinical data.
"By combining clinical expertise with advanced computation, we have realized a long-held vision: a viable path toward early cancer detection via standard blood tests. Being part of this journey has been an extraordinary experience, and I am incredibly proud to contribute to such a significant milestone in the future of diagnostics."
"By combining the power of AI with the rich metabolic information captured by NMR spectroscopy, we have created a tool that can detect pancreatic cancer at its earliest and most treatable stages. Our goal is to bring this technology to clinical practice so that more patients can benefit from timely diagnosis and treatment," said Yu-Ting Chang, M.D., Ph.D., a professor of internal medicine (gastroenterology and hepatology) at National Taiwan University.
To see article on Medical Xpress: https://medicalxpress.com/news/2026-03-ai-blood-early-pancreatic-cancer.html