NTUH unveils AI blood model to detect early-stage pancreatic cancer

Taipei, Feb. 26 (CNA) National Taiwan University Hospital (NTUH) said Thursday it has developed a high-performance diagnostic model capable of detecting precancerous lesions and early-stage pancreatic cancer using just 110 microliters of blood serum, with an accuracy rate as high as 90 percent.

The model, named PanMETAI, was developed through collaboration between NTUH and Academia Sinica, according to an NTUH news release issued Thursday.

The approach integrates artificial intelligence (AI) with nuclear magnetic resonance (NMR) analysis via a liquid biopsy to create a screening platform that is "highly stable and globally scalable," the hospital said.

The findings were published on Feb. 13 in the international journal Nature Communications in a paper titled "PanMETAI -- a high performance tabular foundation model for accurate pancreatic cancer diagnosis via NMR metabolomics."

Chang Yu-ting (張毓廷), a professor of internal medicine at NTUH who co-authored the paper, said at a media event that the greatest difficulty in curing pancreatic cancer in the past has been "discovering it too late."

Pancreatic cancer is often diagnosed at an advanced stage because early symptoms are not obvious and effective screening methods are lacking, and the five-year survival rate is about 13 percent, Chang explained.

Chang said multiple studies have found that pancreatic cancer triggers "a series of progressive metabolic and soft tissue changes" that "offer the potential for early detection of pancreatic cancer."

Advances in big data and AI technology enabled the team, after years of work, to integrate the technology into a platform that requires only a blood sample for precise detection.

Chang explained that the research team used a highly standardized NMR-based metabolomics analysis platform to obtain up to about 260,000 metabolic signals from each 110-microliter serum sample and then applied a deep learning model to systematically extract key features associated with pancreatic cancer.

"This method can comprehensively reflect the overall metabolic changes of pancreatic cancer from precancerous lesions to early-stage cancer, significantly enhancing early risk identification capability," Chang said.

He added that in an independent blind test dataset at NTUH, PanMETAI achieved an overall predictive performance of 0.99, with sensitivity of 93 percent and specificity of 94 percent.

Chang said the technology "is not limited to pancreatic cancer," noting that the team has begun applying the multi-cancer detection platform for detecting stomach cancer, colorectal cancer and liver cancer.

He described preliminary results as "pretty ideal," and that the approach is expected to become an early screening tool across multiple cancers types.