graph TD
    subgraph Preprocessing
        A[DICOM Loader] --> B[dicom2nifti or SimpleITK]
        B --> C[PNG Conversion]
        C --> D[OCR Engine]
        D --> E[PDF Text Extractor]
    end

    subgraph ImageProcessing
        F[TotalSegmentator] --> G[Segmentation Masks]
        G --> H[Constraint Mapper]
    end

    subgraph NLP_and_Symbolic
        E --> I[Biomedical NLP]
        I --> J[NER + Parsing]
        J --> K[State Emitter]
        H --> K
        K --> L[Affinity Graph]
    end

    subgraph Inference
        L --> M[Constraint Lattice]
        M --> N[WildCore Security]
        N --> O[Phi-2 or Gemma-2B]
        O --> P[Symbolic Output]
    end

    subgraph ControlEvaluation
        P --> Q[State Vector & Metrics]
        Q --> R[Audit Logger]
        R --> S[Human-in-the-Loop]
    end

Loading

Remote OHIF Access

The :mod:msk_io.retrieval package provides helpers for working with web-based viewers. OHIFCanvasExtractor captures rendered frames using a headless browser. DICOMStreamSniffer monitors network traffic to download the original DICOM payloads. RemoteDICOMLoader combines both methods and is triggered when remote_url and auth_token are set in PipelineSettings.

Local Chromium Setup

Some utilities rely on a headless Chromium browser. Run ./bootstrap_chromium.sh to ensure a compatible binary is available. The script checks for a system installation and unpacks resources/chromium/chromium.tar.xz when needed. A Python helper (bootstrap_chromium.py) provides the same behavior for restricted environments.

Automated Bootstrap

For air‑gapped execution bootstrap_pipeline.py orchestrates the entire setup. It validates the local Chromium archive, extracts it through bootstrap_chromium.sh and runs the pipeline with MSK_REMOTE_URL and MSK_AUTH_TOKEN from the environment. Once resources/chromium/chromium.tar.xz is present, invoke it as:

MSK_REMOTE_URL="https://nbia.cancerimagingarchive.net/viewer/?study=..." \
MSK_AUTH_TOKEN="eyJhbGciOiJI..." \
python bootstrap_pipeline.py

Quickstart

The following steps reproduce the offline pipeline described in the project whitepaper. They assume a Linux environment:

1. Create and activate a virtual environment:

   python3 -m venv venv
   source venv/bin/activate

2. Install dependencies. Prefer the editable install but fall back to the minimal list when the optional extras are unavailable:

   pip install -e .[dev] || \
   pip install pydantic-settings prometheus-client httpx fastapi numpy pydicom lmdb pillow nibabel pdfminer.six pyppeteer

3. Place a compatible chromium.tar.xz inside resources/chromium/ if the system has no network access. Then bootstrap the browser with:

   ./bootstrap_chromium.sh  # or `python bootstrap_chromium.py`

4. Run the test suite to confirm the environment works:

   ./run_tests.sh

5. Export the remote OHIF url and token and launch the pipeline:

   export MSK_REMOTE_URL="https://nbia.example/viewer/123"
   export MSK_AUTH_TOKEN="<jwt-token>"
   python bootstrap_pipeline.py

SMS Token Utilities

The repository includes sms_module.py which implements a basic SMS token flow. Use python sms_module.py send <phone> <token> to send an encoded token and python sms_module.py receive <message> to validate incoming messages.