How joseberlin7/3d-mri-segmentation works
This repository implements a BraTS 3D medical-image segmentation workflow: subject volumes are loaded and transformed, then a SwinUNETR model is trained and evaluated, with metrics and checkpoints produced. Azure ML submission is an optional execution path; the README also identifies notebooks for metric and visualization review. The main call relationships are supported by the training and submission excerpts; notebook internals were not sampled.
Main entry points: train.py Training entry · data_loaders.py Data loaders · model_code.py SwinUNETR · submit_job.py Azure submission
Components 14
Data preparation
- BraTS dataset
src/data_loaders.py - Transforms
src/data_loaders.py - Data loaders
src/data_loaders.py
Model training
- Training entry
src/train.py - SwinUNETR
src/model_code.py - Train loop
src/train.py - Validation metrics
src/train.py
Monitoring and execution
- TensorBoard logs
src/train.py - Model checkpoints
src/train.py - Azure submission
src/submit_job.py
Review
- Metric visualization
Notebooks/dev_note.ipynb
Other components
- Researcher
- BraTS volumes
- Azure ML
Connections 16
- Researcher → Training entry: starts training
README.md - BraTS volumes → BraTS dataset: provides volumes
src/data_loaders.py - BraTS dataset → Transforms: transforms samples
src/data_loaders.py - Transforms → Data loaders: feeds batches
src/data_loaders.py - Training entry → Data loaders: requests loaders
src/train.py - Training entry → SwinUNETR: creates model
src/train.py - Training entry → Train loop: calls training
src/train.py - Data loaders → Train loop: supplies batches
src/train.py - SwinUNETR → Train loop: trains model
src/train.py - Train loop → Validation metrics: validates periodically
src/train.py - Train loop → TensorBoard logs: writes scalars
src/train.py - Validation metrics → Model checkpoints: saves weights
src/train.py - Researcher → Azure submission: requests job
README.md - Azure submission → Azure ML: submits job
src/submit_job.py - Azure ML → Training entry: runs training command
- TensorBoard logs → Metric visualization: metrics reviewed
README.md