Nodes
3
ingest · classify · emit
Tissue,
classified.
A Camelyon16 ONNX whole-slide pipeline. Ingest gigapixel pyramidal TIFFs. Classify against a packaged tumour model. Ship signed CSV per slide. Three typed nodes for a benchmark-grade classifier.
Model
ONNX
portable · signed
Source
Camelyon16
canonical WSI benchmark
Output
CSV
HIPAA-grade · signed
Three nodes,
a benchmark-grade classifier.
Camelyon16 is the canonical whole-slide-image benchmark for tumour detection in lymph-node sections. Most published pipelines run on bespoke Python notebooks that no clinical IT department will deploy.
This pipeline does the same job in three signed Protor nodes. The slide is read as a pyramidal TIFF (the gigapixel format every clinical scanner emits). It is fed tile by tile to an ONNX model packaged with the pipeline. The classifier emits a CSV row per ROI with probability, score, and a hash of the input tile.
Same engine as the rest of the site. Same evidence chain. Same audit story.
Three nodes,
one .viow file.
From E2E Tests/Pathology/camelyon16-onnx-tumor.viow.
ingestimage.PyramidalTiff.Subscriber
→
classifyclassification.Onnx
→
emitcsv.Publisher
The Pathology folder ships four sibling workflows: a benchmark, a full pipeline, an ONNX-tumour classifier (this one), and an ONNX tissue-detection variant. All share the same three-node skeleton with model swaps.
Four workflows,
same skeleton.
| Workflow | Nodes | Purpose |
|---|---|---|
camelyon16-onnx-tumor.viow | 3 | This page — tumour classification head. |
camelyon16-onnx-classify.viow | 3 | Generic ONNX classifier — swap the model file. |
camelyon16-onnx-tissue-opt.viow | 3 | Tissue-detection variant — fast first-pass to reject empty tiles. |
camelyon16-full-pipeline.viow | 7 | Full pipeline — tissue detect → tumour classify → ROI reporter. |
camelyon16-benchmark.viow | 4 | Throughput benchmark with timing instrumentation. |
Why this matters.
- ONNX, signed, packaged. The model file ships with the workflow. Its hash is on the signed evidence pack. A regulator can verify the exact model that produced any classification.
- Whole-slide, on the workstation. Gigapixel images do not leave the hospital. The whole pipeline runs on the pathologist's machine.