Taiwanese researchers and private enterprises are using artificial intelligence for early detection of some of the deadliest malignancies.

Pancreatic cancer has the highest mortality rate of all major cancers worldwide, with a five-year survival rate of just 13%, according to the American Cancer Society. The disease’s aggressive nature causes it to grow quickly and often spread to other parts of the body, such as the liver, bile ducts, and intestines, as well as major blood vessels, early in its development. As a result, pancreatic cancer is considered inoperable in its advanced stages.

Often asymptomatic in its early stages, pancreatic cancer is exceptionally hard to detect before it has reached an advanced, incurable stage. Unlike lung, colon, or breast cancer, there is no standardized screening protocol for pancreatic cancer, and its location deep within the abdomen makes it difficult to visualize. Existing methods of detection, such as through blood tests and imaging, have limited accuracy, and biomarkers specific to early pancreatic cancer have been challenging to identify.

In Taiwan, pancreatic cancer is the seventh most common malignancy. Its prevalence rose in the country between 2002 and 2013. During that time period, the incidence rate of the disease increased from 4.62 per 100,000 to 6.04 per 100,000, according to a study published in the journal Cancer Medicine in November 2018.

As in many countries, Taiwan’s population aged significantly during that period. Since pancreatic cancer is more common among older adults, this increase in incidence rate is somewhat foreseeable. In addition, Taiwan experienced an increase in obesity rates, leading to greater insulin resistance and inflammation, both of which are linked to an increased risk of pancreatic cancer.

It is against this backdrop that researchers at National Taiwan University Hospital (NTUH) have made a potential breakthrough in diagnosing the disease. In January, NTUH’s pancreatic cancer team announced that they had developed an AI-assisted model for interpreting computed tomography (CT) scans. The advanced system is capable of identifying tumors smaller than 2 cm – lesions that often evade conventional diagnostic methods. Moreover, the model can accurately differentiate between cancerous and non-cancerous pancreatic tissues, enhancing diagnostic precision.

The NTUH team trained the AI algorithm with images from more than 3,000 cases at the hospital and tested it with nationwide clinical data. The sensitivity rate for detecting pancreatic tumors 2 cm or smaller was roughly 80%, while image interpretation can be completed in one minute, National Taiwan University Hospital Department of Internal Medicine Professor Liao Wei-chih told reporters in January.

In a November 2023 article, the Radiological Society of North America (RSNA) lauded the NTUH team’s diagnostic tool, emphasizing its use on whole-body CT scans of automated pre-processing segmentation, or identification and outlining of the pancreas.

“Automation of this process represents an important advance in AI evaluation of pancreas imaging, as the pancreas borders multiple organs and structures and varies widely in shape and size,” the RSNA observed.

The AI-assisted diagnosis system appears to have strong commercial potential. It has received a medical license from Taiwan’s Food and Drug Administration (FDA), a “breakthrough device” designation from the U.S. FDA, and multiple patents in both countries.

However, experts say that the ultimate success of NTUH’s project and others in a similar vein will depend on the healthcare and reimbursement systems of the respective countries where it is sold.

In Taiwan’s case, “with a smaller population, the demand for disease screening is lower than in larger countries,” notes Janiece Kuo, an industry analyst at the semi-governmental Market Intelligence & Consulting Institute (MIC). “However, Taiwan’s universal healthcare system provides coverage, so AI developers typically focus on healthcare screening services eligible for reimbursement.”

Taiwan’s National Health Insurance (NHI) is a single-payer, universal healthcare system that covers most medical services. AI-based healthcare tools must be approved for inclusion to be eligible for reimbursement. Appropriately assessing certain AI technologies’ eligibility is a challenge – not only is the reimbursement system having difficulty keeping up with AI’s rapid development, AI, as an intangible and versatile tool, lacks an exact metric to designate a set value or price for reimbursement.

The future development of this AI system will depend on Taiwan’s future public health policy, Kuo says, adding that large Taiwanese hospitals may license their AI technology to companies for further commercialization.

Further applications

In addition to advancements in pancreatic cancer detection, Taiwan is leveraging artificial intelligence to enhance early diagnosis of other prevalent malignancies. Notably, in July 2024, V5med – a subsidiary of the Taiwanese semiconductor optical inspection equipment giant V5 Technologies – partnered with AstraZeneca Taiwan to promote AI-assisted lung cancer screening.

Given that malignant lung tumors are the leading cause of cancer deaths in Taiwan, the collaboration aims to improve screening quality for high-risk populations and facilitate early intervention treatments.

“This collaboration aims to aid physicians in detecting early-stage lung cancer lesions that may be challenging to identify, coupled with international risk prediction models,” the two companies said in a July news release.

V5Med and AstraZeneca say that their AI-assisted diagnostic tool has a detection sensitivity of more than 95%. Before the introduction of this tool, low-dose computed tomography (LDCT) – the most widely used diagnostic tool for lung cancer screening – only showed a detection sensitivity ranging from 80% to 93% for early-stage lung cancer. Within a year of the AI-assisted tool’s launch in late 2022, the product had been adopted by about 20 hospitals, the companies added.

Meanwhile, in Taichung, China Medical University Hospital (CMUH) has developed an AI-powered “intelligent antimicrobial system” that rapidly identifies drug-resistant bacteria, cutting diagnostic time from 72 hours to one hour.

Deployed in 2022, the system was found to reduce patient mortality by 25% and extends to diagnosing cancer, Parkinson’s disease, heart attacks, and strokes. CMUH’s AI models, supported by Microsoft Azure, are integrated into hospital workflows, often speeding up treatment and improving outcomes. The hospital says it aims to expand the models globally, prioritizing patient privacy while leveraging cloud-based AI tools for healthcare transformation.

Taiwan has also developed AI-supported tools for detecting colon cancer, which is the most common malignancy in the country. In September 2022, Cathay General Hospital and the startup aetherAI announced that they had developed an AI-enabled system able to detect polyps and tumors during colonoscopies with 96% accuracy. Standard colonoscopies, where detection rates vary significantly based on the skill and experience of the endoscopist, typically detect polyps and tumors with 85% to 90% accuracy.

The AI detection system for large intestine polyps acts as a “second pair of eyes” for doctors during colonoscopies, enabling instant detection of polyps. The technology has been shown to improve adenoma detection rates and reduce colorectal cancer risk. Already licensed as a medical device in Taiwan, it is in use at hospitals and health examination centers across the country.

Looking ahead, Kuo sees growing opportunities for Taiwan at the intersection of AI technology and healthcare. She notes the country’s strong medical expertise, which provides a solid foundation for developing AI healthcare applications in healthcare. Additionally, because the NHI has accumulated extensive health data over many years, there are ample opportunities for related research work.

“This integration of high-quality healthcare and long-term data management gives Taiwan a distinct advantage in developing AI-based screening tools,” Kuo says.