AI and molecular testing offer new methods of detecting pancreatic cancers before they become more advanced
Recent precision medicine research shows that deadly pancreatic tumors can be detected earlier than previously possible using molecular testing and artificial intelligence (AI) analysis of radiographic imaging. Pancreatic cancer is one of the most deadly types of cancer, making it imperative to detect as early as possible. New personalized medicine advances will soon enable oncologists to act earlier in treating this type of cancer.
The first major development in early pancreatic cancer recognition came out of a large, multicenter study led by the University of Pittsburgh (Pitt) School of Medicine. This research used molecular markers to identify if pancreatic cysts were cancerous, and the new test developed appears to have exceeded the diagnostic effectiveness of existing methods.
This research focused on the 15% of people who will develop cysts on their pancreas during their lives. A small percentage of these cysts transform into pancreatic cancer. Current guidelines recommend imaging of these cysts to determine if they are cancerous. Techniques used include ultrasound, computerized tomography (CT) scans, magnetic resonance imaging (MRI), and, sometimes, positron emission tomography (PET) scans.
However, Pitt researchers have developed a molecular test, called PancreaSeq, to identify cancerous pancreatic cysts. An article recently published in Gastroenterology outlines how the test identified pancreatic cancer early at the molecular level.
Sequencing Pancreatic Cyst Fluid for Classification
![](https://precision-medicine-institute.com/wp-content/uploads/Aatur-Singhi-MDPhD-Pathology-UPMC.jpg)
PancreaSeq sequenced 22 pancreatic cyst-associated genes, according to the article in Gastroenterology. The results were correlated with endoscopic ultrasound (EUS) findings, ancillary studies, current pancreatic cyst guidelines, follow-up, and expanded testing of postoperative specimens.
“Based on the results of this study, molecular testing of pancreatic cysts is poised to enter international consensus guidelines for the diagnosis of pancreatic cysts and early detection of pancreatic cancer,” said co-senior author Aatur Singhi, MD, PhD, in a University of Pittsburgh Medical Center (UPMC) press release. “Our hope is that PancreaSeq will not only improve early detection of pancreatic cancer but also avoid overtreatment and unnecessary surgery of non-cancerous cysts.”
Singhi sees the ability to detect pancreatic cancer earlier as a key tool in improving outcomes. “The only way we can improve outcomes for pancreatic cancer is to find better treatments or detect it earlier,” he explained. “Our focus has been addressing both of these issues, and especially improving early-stage detection efforts.”
Using AI in Imaging for Pancreatic Cancer
Identifying if a pancreatic cyst is cancerous is a very important step in improving the recognition of pancreatic cancer. A perhaps even more important problem, however, is being able to detect pancreatic cancer on imaging while the tumors are very small. New research out of Taiwan has advanced a solution for this problem.
Researchers from the National Taiwan University have developed a new AI program that improves early recognition of pancreatic cancer. In this effort, the researchers combined five deep-learning models to identify pancreatic cancers using CT scan imaging.
The new AI technology included several important technological advances, including the ability to automatically identify the pancreas on a CT scan image. This is significant because the pancreas varies widely in shape and size, and it touches many organs. Until now, identifying the pancreas using AI has been technically difficult.
More importantly, however, the study, recently published in Radiology, showed that the AI program identified pancreatic cancer as accurately as a radiologist. “The performance of the deep learning tool seemed on par with that of radiologists,” explained Weichung Wang, PhD, professor at National Taiwan University and director of the university’s MeDA Lab, in a Radiological Society of North America (RSNA) article. “Specifically, in this study, the sensitivity of the deep learning computer-aided detection tool for pancreatic cancer was comparable with that of radiologists in a tertiary referral center regardless of tumor size and stage.” Wang was the senior author of the study and helped lead the development of this new tool.
While the AI program successfully identified pancreatic cancer on imaging, it was also able to identify tumors smaller than 2cm 75% of the time. This is a significant improvement over radiologists not assisted by AI, who only have a 60% success rate in identifying these smaller tumors. Catching pancreatic tumors while they are small improves the likelihood that they can be successfully treated.
The researchers see this new tool as something that can help support doctors in the future. “I think AI can do two things,” said Wei-Chih Liao, MD, PhD, in an interview with Inverse. Liao is a professor of internal medicine at the National Taiwan University and one of the study’s co-senior authors. “The first is to help doctors do what they can, but with less time and energy. The second is to help doctors do what they cannot always do; for example, in our study, detect cancers that are not very visible to humans.”
Both the molecular testing for pancreatic cancer and AI analysis of CT imaging are advancing the early recognition of pancreatic cancer. These new precision medicine-based technologies may soon enable this deadly type of cancer to be treated earlier and more effectively, enabling healthcare leaders to provide their patients with better outcomes.
—Caleb Williams
Related Information:
Pitt-Developed Genetic Test for Pancreatic Cancer Outperforms Current Guidelines
Pancreatic Cancer Detection on CT Scans with Deep Learning: A Nationwide Population-based Study
Study Finds Deep Learning Tool Performance Comparable to That of Radiologists
AI Can Now Detect Pancreatic Cancer Better Than Radiologists