Early detection of pancreatic cancer remains a significant challenge in oncology, as the disease is often diagnosed at advanced stages. Now, researchers have developed a method that analyzes metabolic information from blood using artificial intelligence (AI), offering a rapid and non-invasive way to identify the disease in its initial stages.
A team from Taiwan has developed PanMETAI, an AI-powered platform that analyzes metabolic fingerprints from a simple blood sample to detect pancreatic cancer at its earliest stages, when treatment is most likely to be successful. This breakthrough offers hope for improving outcomes in a cancer known for its aggressive nature and late diagnoses.
The study, conducted by researchers at National Taiwan University Hospital and Academia Sinica, was published in the journal Nature Communications.
Pancreatic cancer is one of the most lethal forms of cancer worldwide, with a five-year survival rate of approximately 13%. The difficulty in early detection contributes to this poor prognosis, as most patients are diagnosed when treatment options are limited.
Current screening methods, including the commonly used tumor marker CA19-9 in blood tests, lack the necessary sensitivity and specificity for early disease detection.
The PanMETAI platform combines artificial intelligence with nuclear magnetic resonance (NMR) metabolomics, a technique that identifies the chemical profile of hundreds of metabolites in the blood. Metabolites are small molecules resulting from the body’s metabolic processes, and variations in these can reflect biological changes associated with the disease.
The analysis requires only 500 microliters of blood serum. From this sample, the system extracts over 260,000 metabolic signals, which are analyzed by an AI model called TabPFN.
The model integrates metabolic profiles with the patient’s age, CA19-9 levels, and the protein biomarker Activin A.
In the Taiwanese patient cohort, the platform achieved an area under the curve (AUC) of 0.99, a measure of diagnostic test performance. This indicates the platform correctly differentiated nearly all cases of pancreatic cancer from those of high-risk individuals in the control group.
To evaluate whether the method performs in other populations, the model was tested in an independent cohort from Lithuania, consisting of 322 participants recruited at Lithuanian University of Health Sciences. In this group, the platform maintained a high performance, with an AUC of 0.93, indicating a consistent ability to identify the disease in populations with different ethnic and geographic backgrounds.
A significant finding of the study is the platform’s ability to detect pancreatic cancer in its early stages (stages I and II).
NMR metabolomics significantly contributed to increased sensitivity for early detection, identifying subtle metabolic changes that occur before the disease becomes clinically evident.
These changes included a decrease in HDL cholesterol and glutamine, as well as an increase in lactic acid, glucose, and glutamic acid levels.
Researchers found that the platform could function effectively even when trained with small datasets. The system maintained approximately 90% accuracy even when only 50 cases were used to train the algorithm, suggesting the method could be used in hospitals or research centers that do not have large clinical databases.
The study authors believe PanMETAI could become a rapid, non-invasive, and relatively accessible screening tool used in medical practice to identify individuals at increased risk of pancreatic cancer, who may require further investigation.
They emphasize that integrating machine learning methods with clinical data and a complex metabolic spectrum opens new possibilities for early cancer detection through standard blood tests, demonstrating how fundamental cancer research and artificial intelligence can work together to translate laboratory discoveries into diagnostic applications.
The team’s goal is to introduce this technology into medical practice to allow for earlier diagnosis of pancreatic cancer and prompt initiation of treatment.
