How zhangxd0530/ms-dsa-net works
MS-DSA-NET is a 3D focal cortical dysplasia lesion-segmentation application for paired T1 and FLAIR MRI. Its documented user workflow prepares subject images, registers them to the MNI template, runs the pretrained segmentation model, and saves a lesion map for viewing. The graph preserves the preprocessing, image transforms, network, and lesion post-processing stages, plus evaluation utilities. The inference entry point is described in the README but its source was not sampled, so workflow links rely on that documentation; internal inference details are omitted.
Main entry points: seg_fcd_test.py Test inference · ms_dsa_net.py MS-DSA-NET · preprocess_data.py FSL registration · raw T1 + FLAIR MRI
Components 21
Image preparation
- T1 + FLAIR MRI
inputs/raw - FSL registration
preprocess_data.py - MNI template
inputs/MNI152_T1_1mm.nii.gz - Registered MRI
preprocess_data.py - Input transforms
get_transforms.py
Segmentation
- Test inference
seg_fcd_test.py - Model factory
get_model.py - MS-DSA-NET
networks2/ms_dsa_net.py - CNN and attention blocks
networks2/conv_blocks.py - Prediction transforms
get_transforms.py - Lesion map
seg_fcd_test.py - Pretrained weights
- Saved results
seg_fcd_test.py
Model support
- Lesion post-processing
utils2/utils_common.py - Segmentation evaluation
utils2/utils_common.py - Segmentation losses
get_loss.py - Training transforms
get_transforms.py - GridMask augmentation
utils2/gridmask.py - Learning-rate schedules
utils2/lr_scheduler.py
Other components
- Research user
- MITK viewer
Connections 20
- Research user → T1 + FLAIR MRI: provides
README.md - T1 + FLAIR MRI → FSL registration: registers
README.md - MNI template → FSL registration: guides registration
preprocess_data.py - FSL registration → Registered MRI: writes
preprocess_data.py - Registered MRI → Test inference: loads
README.md - Test inference → Input transforms: prepares images
- Test inference → Model factory: selects model
- Model factory → MS-DSA-NET: constructs
get_model.py - MS-DSA-NET → CNN and attention blocks: uses blocks
networks2/ms_dsa_net.py - Pretrained weights → Test inference: loads weights
README.md - Test inference → MS-DSA-NET: predicts lesions
README.md - Test inference → Prediction transforms: post-processes prediction
- Test inference → Lesion post-processing: cleans prediction
- Test inference → Lesion map: produces
README.md - Lesion map → Saved results: saves
README.md - Saved results → MITK viewer: viewed in
README.md - Segmentation evaluation → Lesion map: evaluates
- Segmentation losses → MS-DSA-NET: scores training output
- Training transforms → GridMask augmentation: may augment with
- Learning-rate schedules → Segmentation losses: schedules optimization