What if AI could find a warning sign in a scan a patient has already had, and give their doctor a reason to look more closely?
Two research projects explore that possibility: one looks at a serious complication of lung-cancer treatment, the other at signs of pancreatic cancer before a tumour is visible. Both are promising, but neither is a new standard of care. These studies come from US researchers; this UK AI Good News update does not describe an NHS rollout.
Could a CT scan flag a lung-treatment complication earlier?
Researchers led by the University of Texas MD Anderson Cancer Center developed a model called CIPHER. It analyses CT scans taken before immunotherapy in people with non-small cell lung cancer, looking for patterns associated with a higher risk of immune checkpoint inhibitor-induced pneumonitis: potentially serious lung inflammation.
The useful idea is not to add an entirely new scan. It is to investigate whether information already in baseline imaging might help doctors identify who needs closer monitoring. The researchers tested the model against patient data from their own centre and a separate Johns Hopkins cohort.
That is a research result, not a clinical decision tool. The study is a preprint that has not been peer-reviewed, and the authors say prospective validation is needed before clinical use. We do not yet know whether acting on its predictions improves patient outcomes.
Source: CIPHER research preprint ↗
Searching for pancreatic cancer before a tumour is visible
At Mayo Clinic, researchers developed REDMOD to analyse routine abdominal CT scans for subtle pancreatic tissue changes in people later diagnosed with pancreatic cancer.
Across the retrospective research, the team analysed nearly 2,000 scans. In an independent test set of 493 scans, REDMOD identified 73% of prediagnostic cancers at a median of around 16 months before diagnosis. These scans had originally been interpreted as normal.
That does not mean REDMOD finds 73% of all pancreatic cancers in the general population. It is a result from a selected retrospective test set, and real-world performance—including false alarms—still needs to be evaluated.
Mayo Clinic says the team is moving into prospective clinical testing through AI-PACED to study how AI-guided detection might work in care for patients at elevated risk. REDMOD is an investigational tool, not an approved pancreatic-cancer screening programme; earlier signals have not yet been shown to improve survival.
Sources: Mayo Clinic: REDMOD results and AI-PACED ↗ · Published research abstract ↗
Finding what people struggle to see
AI does not need to replace the person reading a scan. It could become another pair of eyes, analysing patterns that are difficult to evaluate manually and flagging something worth checking.
Then the clinician decides what happens next. That promise depends on careful testing: a useful signal must hold up in real practice and help people, not simply look impressive in a study.
AI finds the signal. Humans make the decision.
No AI for the sake of AI.
Less hype. More real-world AI progress.