A new artificial intelligence (AI) model developed by researchers at Penn Medicine can identify hidden signs of disease and potential future health risks in CT scans, even when the scans were originally performed to investigate unrelated medical conditions. The findings, published in the peer-reviewed journal NPJ Digital Medicine, suggest that routine medical imaging could provide doctors with a broader understanding of a patient’s health.
The model, named Percival, examines patterns in medical images that may be difficult for clinicians to identify through conventional analysis. Unlike AI systems designed to detect specific conditions, Percival can analyse imaging data more broadly, identifying possible health concerns without being directed to search for a particular disease.
Researchers believe the technology could improve early disease detection and support more personalised healthcare by extracting additional clinical information from scans that patients have already undergone.
“Our teams have developed something that learns broadly from imaging data and uncovers connections between scans, lab values, people’s physiology, and clinical outcomes to potentially give more accurate assessments of health, or even ring alarm bells early,” said senior author Walter Witschey, PhD, an associate professor of Radiology and the associate vice chair of Research IT.
“This is work that points toward a future in which medical imaging becomes a foundation for more personalized and proactive healthcare.”
Percival trained using nearly 400,000 CT scans
Percival was developed through collaboration between several research teams at Penn Medicine using information from the Penn Medicine BioBank, a large repository of patient health data collected across the University of Pennsylvania Health System.
The BioBank contains information from more than 350,000 individuals gathered over approximately two decades. Researchers used roughly 400,000 CT scans from around 50,000 patients to train and refine Percival, which remains in the preclinical testing stage.
The model’s name references Percival, one of the knights of King Arthur’s Round Table, while also honouring the late Eugene Percival Pendergrass, a prominent radiologist and educator at the University of Pennsylvania.
According to the researchers, access to extensive and diverse medical datasets is important for developing AI systems that can perform reliably across different patient populations.
“Research in AI has shown that training on larger, more diverse datasets can improve performance, and this can provide a stronger foundation for adapting a model to a hospital or patient population almost anywhere,” said lead author Cameron Beeche, a PhD candidate in Bioengineering at the University of Pennsylvania.
The researchers are particularly interested in expanding the use of opportunistic screening, a process in which medical professionals identify additional health concerns while examining scans performed for another purpose.
For example, a CT scan ordered to investigate pneumonia could also contain early indications of cancer or other underlying conditions that might otherwise remain undetected.
“We’re hoping to elevate ‘opportunistic screening,’ where results are found beyond the original purpose of a scan, such as finding the early traces of cancer on a scan that was maybe assessing pneumonia,” Witschey said.
“We believe the totality of what’s in a CT scan can be made much more useful with help from a model like this.”
AI model identifies chronic diseases and predicts complications
During performance evaluations, Percival demonstrated the ability to identify imaging patterns associated with several chronic health conditions, including atrial fibrillation, a type of irregular heartbeat, and chronic kidney disease.
Researchers also compared Percival with TotalSegmentator, another AI model used in medical imaging. The findings showed that Percival could identify conditions such as pancreatic cancer, cirrhosis and bacterial pneumonia more reliably than the comparison model.
Beyond identifying existing diseases, the researchers found that Percival could help predict disease progression and assess the likelihood of major cardiovascular and liver complications.
“Percival can help identify unrecognized disease, predict future progression, and anticipate major cardiovascular and liver complications,” said co-author Charles Kahn, MD, a professor of Radiology.
“Because many patients already have substantial heart or metabolic disease when they enter specialty care, routine imaging may provide an opportunity for earlier recognition and intervention.”
Despite these findings, Percival has not yet been introduced into routine clinical practice. The research team plans to begin further testing in early 2027, with the longer-term objective of making the technology available for regular clinical use.
The developers expect their experience with AInSights, another AI-based radiology system developed at Penn Medicine, to support Percival’s transition towards clinical applications.
Meanwhile, researchers are exploring whether larger datasets and information collected from multiple medical institutions could improve the model’s reliability and predictive capabilities.
Percival’s code and training model specifications have also been made publicly available through an open-source repository, allowing other researchers to examine and potentially build upon the technology.
“Percival benefited enormously from researchers who shared their code, methods, and insights into these types of models,” Witschey said. “We believe strongly in contributing to that same spirit of open science.”
The development adds to Penn Medicine’s growing portfolio of AI-based medical imaging technologies. Alongside Percival and AInSights, its researchers have developed a separate model designed to analyse cardiac MRI scans and identify medical conditions, with testing reporting near-perfect diagnostic accuracy.
While further evaluation is necessary before Percival can be used routinely in hospitals, the study suggests that AI could help doctors obtain more clinical information from existing medical scans. Such advances may eventually support earlier diagnosis, improved risk assessment and more informed treatment decisions without requiring additional imaging solely to investigate previously unrecognised conditions.
Read More :