ML Model Demo - Chest X-Ray Generation

1. Controls & Generation

Limited compute budget, please ping the Endpoint to spin it up if idle. Shouldn't take more than a few minutes.

Checking status...

Endpoint

Model Variant (Cond. vs General)

Guidance (Adherence)

Seed (Reproducibility)

Steps (Quality)

Conditions (Select one or more clinical findings)

No Finding Enlarged Cardiomediastinum Cardiomegaly Lung Opacity Lung Lesion Edema Consolidation Pneumonia Atelectasis Pneumothorax Pleural Effusion Pleural Other Fracture Support Devices

2. Generated Image

Generated image

3. About This Tool

This tool is a demo of a hand-built class-conditioned diffusion model that was trained from scratch to generate medically useful synthetic 128x128 chest X-ray images without exposing patients via memorized content from training data

Guidance: Classifier-Free Guidance scale, higher values mean higher class fidelity, i.e. closer adherence to the conditions checked
Reproducibility: Same seed & settings = same image.
Quality: More inference steps can improve quality but take longer.