Skip to content

quickstart.md

Latest commit

 

History

History
77 lines (58 loc) · 2.41 KB

quickstart.md

File metadata and controls

77 lines (58 loc) · 2.41 KB
Raw

Quick Start

Install the checkout

git clone https://github.com/Liohtml/MedCheck.git
cd MedCheck
uv sync
uv run medcheck serve

Open http://localhost:8080. Upload a DICOM file or ZIP, inspect available studies, select a study and start local analysis. Follow progress, inspect slices and download a report. Local statistics do not generate diagnostic findings and need no cloud account. The web application defaults to metadata de-identification.

Docker

docker build --target lite -t medcheck:lite .
docker run --rm -p 127.0.0.1:8080:8080 \
  -v medcheck-state:/app/.medcheck medcheck:lite

To read server-side folders, additionally mount a source directory and configure the allowed root:

docker run --rm -p 127.0.0.1:8080:8080 \
  -v "$PWD/scans:/data/scans:ro" \
  -v medcheck-state:/app/.medcheck \
  -e MEDCHECK_DATA_ROOT=/data/scans medcheck:lite

The lite target includes cloud SDKs and local image statistics. The full target adds PyTorch/torchvision for feature extraction; neither downloads or bundles a vision language model. State storage contains sensitive data. See Workbench for retention and deployment settings.

CLI

# Local DICOM file, folder, ZIP or DICOMDIR:
uv run medcheck analyze ./scans \
  --steps ingest,preprocess,ml_analysis,report --deidentify --report json

# Interactive input:
uv run medcheck analyze ./scans --interactive

Use --study-uid when the source contains multiple studies. Reports are written to ./output/ unless --output is specified.

For cloud vision, install the relevant SDK and set its API key in the environment:

uv sync --extra claude
export ANTHROPIC_API_KEY=your_key
uv run medcheck analyze ./scans --model claude \
  --deidentify --allow-cloud-llm --pixels-reviewed --report pdf

Only confirm --pixels-reviewed after inspecting the images and text for identifiers. De-identification and optional OCR cannot guarantee anonymous input. For another provider use --extra openai or --extra gemini; --extra cloud installs all three. See Models for a user-managed local vision server.

Evaluate saved reports

Prepare independently labelled reference cases and run:

uv run medcheck evaluate manifest.json --output evaluation.json

See Evaluation for manifest format, baseline comparisons and limitations. These metrics measure label agreement, not clinical accuracy.