Spleen CT Segmentation in the Browser

A U-Net trained on abdominal CT segments the spleen on held-out patients it has never seen. The model runs on your device with ONNX Runtime Web: no server, and no image ever leaves your browser.

Model prediction Radiologist label

Test patient Slice through the spleen

Prediction Label outline

Slice Dice: –
Inference time: –
Backend: –

Or try your own axial CT slice (PNG/JPG, soft-tissue window)

Loading model…

How it was built

Data: 41 labeled CT volumes from the Medical Segmentation Decathlon (Task 09, spleen), split by patient so no test patient's slices are seen in training. Training, evaluation, ONNX export and deployment of this page all run in GitHub Actions. Metrics are per-patient 3D Dice and HD95 on the held-out patients. Code and model card on GitHub.

Research demo only, not a medical device. Not for diagnosis or clinical use.