gs://…/pathdemo/<case_id>/base.jpg, then click "Check for new cases" below — no redeploy needed.
demo_path_annoatate/ on the server.Not loaded — click "Load model" to read the checkpoint into memory on this machine.
This is a self-supervised feature extractor, not a classifier — it returns an embedding vector per image, not a diagnosis. Useful for similarity search, clustering, or as input to a classifier you train yourself.
Not loaded — click "Load model" to initialize it.
Each foreground tile's hematoxylin channel is segmented independently; nucleus centroids are converted to whole-slide pixel coordinates as tiles are stitched back together.
Scroll to zoom, drag to pan. Hover a dot to see its slide-pixel coordinate — at low zoom, only a sampled subset of cells renders for performance; zoom in to see every one. Click a cell to predict its gene expression below.
The segmentation model's own certainty that it found a real nucleus at each dot — not a measure of tissue or diagnostic significance.
This runs NephroLens's AI models, which need the GPU server. The GPU server is switched off to save cost and is started on request — ask the NephroLens team to switch it on.
Pathology Annotation, DemoPathologyAnnotation and DemoPathReport work as usual.