Multimodal AI for glioma analysis. NVIDIA Academic Grant Program.
| Module | Purpose |
|---|---|
schema.py |
Niche label space (IvyGAP), rung hierarchy, provenance manifest, Q1 audit |
ingest/opensrh.py |
OpenSRH two-channel SRH patch reader (verified byte format) |
render/falsecolor.py |
Beer-Lambert virtual H&E; wavelength sensitivity sweep |
models/detectors.py |
Swin-backboned Mask R-CNN and Mask2Former |
experiments/learning_curve.py |
Pre-registered sample-efficiency ablation |
holder/scaling.py |
Hölder/WTMM exponents, Koopman ladder, zoom-Lyapunov |
topology/persistence.py |
Alpha shapes, persistent homology, niche topology QC |
roi/heterogeneity.py |
Procedural-noise null model for ROI detection |
cli.py |
ingest-opensrh, audit, render-srh, selftest |
python -m radiant_beam.cli selftest
python -m radiant_beam.cli ingest-opensrh /data/opensrh --pilot
python -m radiant_beam.cli audit
python -m radiant_beam.cli render-srh patch.tif --out he.pngCore: numpy, scipy, scikit-image, tifffile, pydicom Math: PyWavelets (Hölder), gudhi (topology), opensimplex / noise (ROI) Deep: torch, torchvision, transformers, torch_geometric Domain: ramanspy
Modules raise ImportError with an actionable message rather than degrading
silently when an optional dependency is absent.
