AI-Powered Blood Test Detects Early Pancreatic Cancer with More Than 90% Accuracy

As a result, most patients are diagnosed late, and the five-year survival rate remains around 13%. Early disease detection has been a major challenge in oncology. A new artificial intelligence (AI) model has been designed to identify early molecular signals of pancreatic cancer before clinical presentation.

Scientists from Academia Sinica (Taipei, Taiwan) and National Taiwan University Hospital (NTU, Taipei, Taiwan) have created PanMETAI, an AI-powered foundation model for early pancreatic cancer screening. The system combines nuclear magnetic resonance (NMR) metabolomic analysis with the TabPFN AI architecture to detect metabolic changes associated with the earliest stages of pancreatic cancer.

PanMETAI analyzes large-scale metabolic datasets generated from blood samples. Using the standardized NMR platform, the model processes up to 260,000 molecular data points per individual to identify subtle metabolic patterns linked to cancer development. Unlike conventional diagnostic methods that rely on single biomarkers, the model evaluates comprehensive metabolic profiles, enabling the detection of molecular changes from precancerous states through early-stage disease.

In validation studies published in Nature Communications, the model demonstrated strong diagnostic performance. In an independent blind test conducted in Taiwan, PanMETAI achieved an area under the curve (AUC) of 0.99, with 93% sensitivity and 94% specificity. In a separate European validation using a Lithuanian cohort, the system maintained high predictive accuracy with an AUC of 0.93, demonstrating consistent performance across diverse populations. Because PanMETAI captures the complete metabolic characteristics associated with pancreatic cancer development, it provides a more comprehensive diagnostic approach than traditional biomarker-based screening.

Researchers believe PanMETAI could become an important screening tool for individuals at high risk of pancreatic cancer. Early identification of metabolic changes may allow clinicians to intervene before the disease progresses to advanced stages. The team also envisions expanding the technology into a multi-cancer early detection platform, applying the same AI-driven metabolic analysis approach to identify additional cancer types and support broader precision medicine strategies.

Related Links:
Academia Sinica
NTU