How yuhangliu98/mred-net works
MRED-Net is a low-dose CT denoising research application. Its principal workflow prepares DICOM scans as normalized NumPy slices, loads paired low-/full-dose data, trains or evaluates a Mamba-based residual encoder–decoder, and reports denoising quality. The README documents the ET2T, Mamba, encoder–decoder, and compound-loss capabilities; source excerpts confirm the entry point, loader, model components, and solver responsibilities. Some architectural links are based on README statements rather than inspected call sites.
Main entry points: solver.py Train and test solver · MREDnet.py MRED-Net · main.py Run configuration · prep.py DICOM preparation
Components 15
CT Data
- DICOM scans
- DICOM preparation
MRED-net/prep.py - Normalized NPY pairs
- CT dataset loader
MRED-net/loader.py - Patch augmentation
MRED-net/data_aug.py
Denoising Model
- MRED-Net
MRED-net/MREDnet.py - ET2T blocks
MRED-net/MREDnet.py - Mamba state-space blocks
MRED-net/MREDnet.py - Residual encoder–decoder
MRED-net/MREDnet.py
Training and Evaluation
- Run configuration
MRED-net/main.py - Train and test solver
MRED-net/solver.py - Compound and feature losses
MRED-net/loss.py - Image quality metrics
MRED-net/measure.py
Outputs
- Checkpoints and figures
Other components
- Researcher
Connections 16
- Researcher → DICOM preparation: prepares scans
README.md - DICOM scans → DICOM preparation: reads scans
MRED-net/prep.py - DICOM preparation → Normalized NPY pairs: writes normalized slices
MRED-net/prep.py - Researcher → Run configuration: configures run
README.md - Run configuration → CT dataset loader: builds data loaders
MRED-net/main.py - Normalized NPY pairs → CT dataset loader: loads paired slices
MRED-net/loader.py - CT dataset loader → Patch augmentation: augments patches
MRED-net/loader.py - Run configuration → Train and test solver: starts training or testing
MRED-net/main.py - Train and test solver → MRED-Net: runs denoising model
MRED-net/solver.py - Train and test solver → Compound and feature losses: uses training losses
MRED-net/solver.py - Train and test solver → Image quality metrics: measures image quality
MRED-net/solver.py - Train and test solver → Checkpoints and figures: saves checkpoints and figures
MRED-net/solver.py - MRED-Net → ET2T blocks: uses local token blocks
README.md - MRED-Net → Mamba state-space blocks: uses state-space blocks
README.md - MRED-Net → Residual encoder–decoder: forms residual architecture
README.md - Train and test solver → Researcher: reports denoising results
README.md