One full-size view at a time — switch below. Drag to move the crosshair, scroll to change slice, or use the slider.
Axial — top-down
Coronal — front
Sagittal — side
3D Render
Together
Window
Zoom100%
Loading volumes into the viewer…
Free-air detection — pneumoperitoneum
Why it was flagged
How the decision is made — each measure scored 0–1
How the detector works
Find every voxel dark enough to be air (air ≈ −1000 HU)
Remove air that sits inside the lungs, stomach and bowel, using the organ
segmentation — this step does most of the work
Group what remains into separate collections (“candidates”)
Score each candidate on the measures below; flag the case if any scores high
What is not yet known. Every case this detector has been scored against is one
where we inserted the free air ourselves, so the volumes and recall above are measured
against our own construction. It has never been tested on a confirmed real perforation.
The clean-scan figure above counts elective abdominal CTs
(MSD Task09 spleen cases) where free air is not expected — presumed clear
from the clinical context of that dataset, not confirmed by a radiologist. A
cohort that size cannot establish a false-alarm rate.
The score below is a weighted sum of the measures, not a probability, and is
deliberately shown without a calibrated confidence because there is no labelled
data to calibrate it against.