Replaying old radiographs to ask whether it could have been caught earlier — fifteen years of images before an IPF diagnosis, read again by AI — a clinical trial (ClinicalTrials.gov)
A retrospective observational study of 175 patients diagnosed with idiopathic pulmonary fibrosis, reaching back through the chest radiographs taken before diagnosis to compare the date AI first detected interstitial opacity with the date a radiologist first reported it. The imaging window runs fifteen years, from January 2010 to April 2025, and the registered start and completion dates fall on the same day.
Trial overview (primary data)
- StatusCompleted
- ConditionsIdiopathic Pulmonary Fibrosis, Interstitial Lung Disease
- InterventionsDEVICE: VUNO Med-Chest X-ray
- SponsorChung-Ang University Hospital
- Target enrollment175 participants
- Period2025-04-30 〜 2025-04-30
Key points
- A single-center retrospective observational study comparing, in 175 patients with idiopathic pulmonary fibrosis, the date AI detected interstitial opacity against the date a radiologist reported it.
- The imaging window spans fifteen years, January 2010 to April 2025, reaching back from an April 2025 registry screening.
- Start and completion dates are both 2025-04-30, so the registered duration is zero days, a signature of retrospective work on data that already exists.
- The AI decision uses a score cutoff fixed in advance. The comparator is not reading ability but the finding as documented in the record.
- This describes the design and aims of the study; effectiveness has not been established.
1A question only a backward look can ask
Idiopathic pulmonary fibrosis stiffens the lungs for reasons that stay unknown, and the record states a median survival of only three to five years after diagnosis. Finding it early and starting antifibrotic therapy may change outcomes, yet in practice the diagnosis is often late.
Chest radiography is cheap and available almost everywhere, but its sensitivity for early interstitial opacity is limited, and time can pass without the finding being written down. What this study asks is not whether the AI is accurate. It asks whether, back then, it could have been caught earlier.
2A trial that lasts zero days
The registered start and completion dates fall on the same day, so the duration comes out as zero. The data already exists. A retrospective study never has to wait for patients, and in exchange it can only ask what the images already taken and the reports already written will support. Set against a median duration of 534 days across the records held here, the speed of this rung and its limits appear together.
3What is compared with what
The cohort is 175 patients diagnosed with idiopathic pulmonary fibrosis at a single center. Anchored to a registry screening in April 2025, the study reaches back across fifteen years of imaging, from January 2010 to April 2025.
The point is the gap between the day the AI caught the opacity and the day a radiologist wrote it down, with exploratory work on which patient characteristics widen that gap and whether the trajectory of the AI score relates to mortality.
4It is the record that is being evaluated
What stands on the other side of this comparison is not a radiologist's ability to read, but the finding as it was written into the record. Declining to mention a faint opacity that might one day become fibrosis, on a film taken to answer an acute problem, is not an error in the care of that moment. A backward look deals with the difference that appears only when those moments are laid out again as a series. In a prospective trial, that difference can no longer be measured.
Why it matters
A retrospective design can only ask what past images and past records will support, and in exchange it can measure a difference no prospective trial can reach. Laying documentation out as a series to ask what might already have been visible transfers well beyond imaging, to any field where records accumulate. Defining the comparator as what was documented, rather than as ability, is itself a useful move in evaluation design.
FAQ
Why is the duration zero days?
Is this a study about radiologists missing findings?
Does this establish that the AI is useful?
Sources (primary)
Source: ClinicalTrials.gov (U.S. NIH/NLM, public domain). This site does not provide medical advice. Verify the latest and exact details with the official source. This site is not endorsed or certified by the NIH/NLM.
- ClinicalTrials.gov (study record, original)
- NCT ID: NCT07712952