Penn Medicine AI Percival detects unseen health issues in CT scans, with limits
A new AI model analyzes CT scans to reveal hidden health signals beyond the original test, but findings are early and not yet clinical proof.
What happened
Researchers trained Percival on hundreds of thousands of CT scans and found it uncovers unseen patterns and future health risks beyond the initial reason for the scan, running in the background to pick up what is unseen.
Why it matters
If validated clinically, Percival could support proactive care by flagging early risks; open-source code may invite broader validation and improvements across settings.
The Penn Medicine team developed Percival to analyze CT scans beyond the original testing purpose, aiming to detect signals of disease and forecast future health issues that may not be obvious to human readers. Using data from more than 350,000 people, the model demonstrated the potential to identify patterns linked to chronic conditions such as atrial fibrillation and chronic kidney disease, and it outperformed a similar model in flagging certain cancers and liver issues.
A key goal is to make imaging more useful for proactive care, not just diagnosis. The researchers emphasize openness,Percival’s code and training specs are open source to encourage broader validation and improvement, while continuing to test robustness before regular clinical use.
limitation_per_field_used_is_missing_or_not_confirmed_? Yes, the study notes preclinical testing and relies on data from a single health system; broader validation is required.
What this does not tell us
Results are preclinical and based on data from the Penn Medicine BioBank; generalizability to other populations or settings requires further study.
FOR PEOPLE
Benefits reportedDoctors and patients could gain clearer diagnostic support and earlier interventions.
FOR AI AND ITS OPERATORS
Benefits reportedAI expands data from imaging to inform medical decisions.
These are two separate readings of what the sources describe. Reported claims and risks do not by themselves establish a real-world effect.
Original sources · 1
- An AI model can detect hidden health issues in CT scans ↗Penn Medicine · 2026-10-16
Reporting discovered in United States. Discovery market does not mean the event happened there.
What is your take?
Ask a question, add useful context or share a different perspective. Keep the conversation respectful and grounded.