Neurapedia

Neurapedia is a local, research-only neuroimaging inspection toolkit. It accepts small NumPy volumes in the core installation and provides optional NIfTI/DICOM adapters for real imaging files.

It is not a diagnostic system and does not provide treatment recommendations. Every experimental observation requires qualified human review.

Quick start

Requirements: 64-bit CPython 3.10–3.12 and NumPy. The core workflow runs locally without a server, database, cloud account, or model download.

python -m pip install -r requirements.txt
python -m neurapedia demo --shape 48 48 24 --output neurapedia-report.html

Open the generated HTML file locally. A JSON sidecar is created beside it.

Analyze a NumPy volume:

python -m neurapedia analyze scan.npy --output scan-report.html

Install real imaging readers only when needed:

python -m pip install -e .[imaging]
python -m neurapedia analyze scan.nii.gz --output scan-report.html
python -m neurapedia analyze dicom-study/ --format dicom --output dicom-report.html

The direct launcher also works from a fresh checkout:

python run_neurapedia.py demo

Tests

The shipped tests use the standard library test runner:

python -m unittest discover -s tests -v

Optional development tools are available with python -m pip install -e .[dev].

What the pipeline does

• Validates numeric 3-D and 4-D volumes and rejects non-finite data.
• Loads .npy volumes with the core installation.
• Loads NIfTI and DICOM only through explicit optional adapters; missing or invalid files fail clearly and never become fabricated synthetic scans.
• Generates deterministic synthetic MRI fixtures for demonstrations.
• Produces a disjoint, coarse research atlas with ten named region labels for visualization and testing. It is not a clinical anatomical segmentation.
• Detects robust intensity observations and connected components. It does not assign diagnoses or calibrated clinical probabilities.
• Writes self-contained HTML and JSON research summaries with source metadata and safety warnings.

Optional open-source backends

The deterministic report backend works offline by default. OpenAICompatibleBackend can summarize already-computed structured measurements through a local OpenAI-compatible server such as Ollama or vLLM. Configure the endpoint and model yourself; Neurapedia never downloads a model or sends imaging data automatically.

Open-source imaging model integrations can be added through optional MONAI or nnU-Net adapters, but they require compatible checkpoints, hardware, validation data, and a modality-specific workflow. They are not silently used as a substitute for the deterministic core.

Privacy and safety

Patient labels are kept local. DICOM loading defaults to an anonymous label and does not print DICOM identifiers. The project does not transmit scans. Before using real data, follow your institution’s de-identification, storage, access-control, and review requirements.

Layout

neurapedia/
├── neurapedia/       # package, CLI, loaders, analysis, reports, backends
├── tests/            # standard-library regression tests
├── examples/         # runnable demo
├── pyproject.toml    # package metadata and optional extras
└── run_neurapedia.py # fresh-checkout launcher

License

MIT. See LICENSE.