Penn Medicine's 'Percival' AI reads CT scans for hidden disease beyond the reason they were ordered

Researchers at Penn Medicine (University of Pennsylvania) have developed an AI model called Percival that looks at CT scans for signs of disease and future health risks beyond the reason a scan was ordered, according to a study published in npj Digital Medicine and announced on 16 October 2026.

What the study reports

• Percival was trained on roughly 400,000 CT scans from about 50,000 patients, using data from the Penn Medicine BioBank, which holds information on more than 350,000 people collected over two decades.
• The model found traces of chronic conditions such as atrial fibrillation (an irregular heartbeat) and chronic kidney disease.
• Compared with another AI model, TotalSegmentator, Percival flagged conditions such as pancreatic cancer, cirrhosis and bacterial pneumonia “much more reliably”, according to Penn Medicine.
• The goal is “opportunistic screening” — spotting, for example, early traces of cancer on a scan that was done to check for pneumonia.
• The code and training specifications are open source.

Limitations

• Percival is still being tested preclinically; it is not in routine clinical use. The team hopes to begin testing it early next year with a view to clinical use.
• It was trained on data from a single health system; the researchers say they are studying larger, multi-institution datasets to make the model more robust.
• The summary here is based on the university’s press release; detailed accuracy figures were not included in it. Opportunistic findings can also lead to extra tests and false alarms, which is why prospective validation matters.

This is research news, not medical advice. Talk to your doctor about your own imaging results.