Research demo · not a diagnostic device
Upload a chest X-ray. Three independently-trained, open-source models each score it on their own — shown separately, side by side, with a plain average only where they measure the same thing. No invented ensemble score. No clinical claim.
Two independently-trained DenseNet121 classifiers from the open-source
torchxrayvision library, plus one architecturally-distinct
Vision Transformer fine-tuned by a third party — a genuine cross-check,
not one model wearing three hats.
DenseNet121 · aggregated
Trained on NIH ChestX-ray14 + PadChest + CheXpert + MIMIC-CXR + RSNA (aggregated). 18-label multi-pathology sigmoid output.
DenseNet121 · CheXpert
Same architecture and library, independently trained on CheXpert (Stanford) only — a real second opinion, not a duplicate.
Vision Transformer
ViT-base/16, fine-tuned by a third party on the Kaggle pediatric pneumonia dataset. A different architecture, different training group — the strongest cross-check in the set.
One workflow, deliberately narrow. No model picker, no marketplace, no NLQ.
01
A single chest X-ray, JPEG or PNG, up to 10 MB.
02
All three models score it independently, in parallel.
03
Each model's raw output shown under its own name, never pre-blended.
04
A plain average and agreement note, only where models measure the same thing.
A static preview of how this could price as a product. No checkout here — this pilot doesn't take payment.
Illustrative only — no live billing is wired up in this demo.