How joshykoshy/argus works
Argus is a research pipeline for testing whether frequency-band models improve brain-tumour segmentation under simulated low-field MRI degradation. The main workflow prepares BraTS data, trains segmentation models, evaluates them across degradation conditions, and analyzes segmentation and volume agreement; the repository also includes a degradation-range pilot and 3D visualization. README-level workflow links are documented; several internal module connections are omitted where the supplied excerpts do not establish who calls whom.
Main entry points: results Results and tables · train_v2.py Model training · evaluate_v2.py Test evaluation · prepare.py Array preparation
Components 24
Data preparation
- Data manifest
python/data/manifest.py - Patient splits
python/data/splits.py - Preprocessing
python/data/preprocess.py - Array preparation
training/prepare.py - Slice dataset
python/data/dataset.py
Degradation and models
- MRI degradation
python/physics/degradation.py - Degradation pilot
pilot/pilot.py - Frequency bands
python/physics/decomposition.py - U-Net
python/models/unet.py - Dual-stream U-Net
python/models/dual_stream.py - Model factory
python/models/factory.py
Training and evaluation
- Model training
training/train_v2.py - Test evaluation
evaluation/evaluate_v2.py - Segmentation metrics
python/metrics/evaluation.py
Analysis and validation
- Statistical analysis
python/analysis/stats.py - Morphometrics
matlab - Volume agreement
matlab/agreement_report.m
3D visualization
- Surface meshes
viz3d/meshes.py - Interactive viewer
viz3d/viewer.py - Paper figures
viz3d/paper_figures.py
Other components
- Researcher
- BraTS scans
- Results and tables
results - 3D case views
figures
Connections 29
- Researcher → Array preparation: prepares data
README.md - BraTS scans → Preprocessing: provides scans
README.md - Data manifest → Array preparation: identifies patients
- Patient splits → Array preparation: selects cohorts
training/prepare.py - Preprocessing → Array preparation: processes patients
training/prepare.py - Array preparation → Model training: writes training arrays
README.md - Slice dataset → Model training: supplies slices
- MRI degradation → Model training: augments scans
README.md - Frequency bands → Model factory: provides frequency bands
python/models/factory.py - U-Net → Model factory: instantiated by
python/models/factory.py - Dual-stream U-Net → Model factory: instantiated by
python/models/factory.py - Model factory → Model training: builds models
training/train_v2.py - Researcher → Model training: trains models
README.md - Degradation pilot → Results and tables: reports pilot metrics
pilot/pilot.py - Model training → Results and tables: saves checkpoints
README.md - Results and tables → Test evaluation: provides checkpoints
README.md - Test evaluation → Segmentation metrics: computes measures
- Test evaluation → Results and tables: writes evaluation metrics
README.md - Researcher → Test evaluation: evaluates test set
README.md - Results and tables → Statistical analysis: provides metrics
README.md - Statistical analysis → Results and tables: writes statistics
README.md - Results and tables → Morphometrics: provides predictions
README.md - Morphometrics → Volume agreement: reports agreement
- Results and tables → Surface meshes: provides segmentations
- Surface meshes → Interactive viewer: builds surfaces
viz3d/viewer.py - Surface meshes → Paper figures: builds surfaces
viz3d/paper_figures.py - Test evaluation → Interactive viewer: exports prediction NIfTIs
viz3d/viewer.py - Interactive viewer → 3D case views: writes case views
viz3d/viewer.py - Researcher → Interactive viewer: renders comparisons
README.md