Skip to content

API.md

Latest commit

 

History

History
69 lines (53 loc) · 2.43 KB

API.md

File metadata and controls

69 lines (53 loc) · 2.43 KB
Raw

Inference API (deploy summary)

Full contract lives in the dev repo at inference/API_CONTRACT.md. This page covers what operators need for the pull-only stack.

Endpoints

Public entry (browser / remote): HTTPS via Caddy on https://breader.medinformatics.eu/ (optional HTTP Basic Auth).

URL Purpose
https://127.0.0.1:8443/healthz Liveness + embedding_backend, retrieval_kinds, CXAS status
https://127.0.0.1:8443/metadata Artifact profile, model dirs, retrieval options
https://127.0.0.1:8443/predict DICOM or PNG → heads + neighbors
https://127.0.0.1:8443/docs OpenAPI / Swagger
https://127.0.0.1:8443/ Gradio UI

Orthanc on the Docker network still uses plain http://inference:7860 (no TLS/auth). Host port 7860 is bound to localhost only for debugging.

POST /predict

Content-Type: multipart/form-data

Field Required Default Notes
file yes — PNG or DICOM
top_k no 3 Neighbor count (1–10 in UI)
retrieval_embedding no ark-1376 ark-1376, elixr-768, or elixr-contrastive-img
include_neighbor_thumbnails no false Base64 JPEG per neighbor (Orthanc PDF uses this)

Classifier heads (always Ark 1376-D embeddings):

Key Summary label
q2a Q2A — any small opacity
q3a Q3A — any pleural abnormality
normal_vs_not Strict normal vs not-normal
pleural_calc_any Pleural calc — screen (any site)
pleural_calc_face Pleural calc (face)
profusion_0_3 Small opacity profusion (0–3)
profusion_full_category Full ILO small-opacity category
small_opacities_multiclass Small opacity type (pure cohort)
large_opacity_stage Large opacity stage

Neighbor search uses the selected retrieval space (default Ark). Google spaces load TensorFlow on first use and need extra RAM.

Example (through Caddy)

Use -k for a self-signed origin cert. Add -u user:pass when CADDY_BASIC_AUTH=true:

curl -sk -u breader:'your-long-password' \
  -X POST https://breader.medinformatics.eu/predict \

-F "file=@study.dcm"
-F "top_k=3"
-F "retrieval_embedding=ark-1376" | jq .


Localhost debug (no auth; not published on the LAN):

```bash
curl -s -X POST http://127.0.0.1:7860/predict \
  -F "file=@study.dcm" \
  -F "top_k=3" \
  -F "retrieval_embedding=ark-1376" | jq .