Input gate
X-Ray Validator
The gate at the front. It confirms that what arrived is a chest radiograph before anything else runs, so every result the site produces is a result about an actual chest X-ray.
Purpose
If the image is not a chest radiograph the pipeline stops here and says so, instead of producing a pneumonia score about something that is not a chest. Everything downstream assumes it is looking at a chest X-ray; this is where that assumption is checked rather than trusted.
What the accuracy figure does and does not say
It got every one of 25 images right — 15 chest radiographs and 10 non-radiographs, including a photograph of X-ray film. That is a spot check, not a test-set metric, and the distinction matters: twenty-five images is a sanity check that the gate is wired up and pointing the right way, not evidence of how often it is correct. We have not run a large-scale evaluation of this model, and until we do, treat it as a filter against obvious non-radiographs rather than as a guarantee.
What it deliberately does not check: it does not separate a frontal view from a lateral one, and it does not separate an adult chest from a child's. Both pass, and both sit outside the distribution the classifiers were trained on — a result from either is worth less than the number beside it suggests.
Technical details
- Input
- 1 × 3 × 224 × 224
- Threshold
- 0.5
- Decision rule
not_xray if sigmoid(logit) >= 0.5 else chest_xray- Spot check
- 100% of 25
- Large-scale evaluation
- not run