Breast MRI Tissue Segmentation (nnU-Net)
This repository contains the trained weights as an attachment to the Master's Thesis: Characterization of Breast Tissue in Dynamic MR Scans.
The models were developed using the nnU-Net framework.
🚀 Main Pipeline Model
The the final model for fibroglandular tissue and vessel segmentation is FGT-Vessel-Net-v1.
📂 Overview of Models & Experiments
| Model Name | Description |
|---|---|
Dataset001_DukeFirstTry |
Initial exploratory model |
Dataset002_VesselsPreliminary |
Initial model for vessel GT refinement |
Dataset003_VesselsPreliminary02 |
Iterative refinement model (Iteration 1) |
Dataset004_VesselsPreliminary03 |
Iterative refinement model (Iteration 2) |
Dataset005_Vessels_1st_sub |
Vessel-Net |
Dataset006_FGT_0th |
FGT-Net |
Dataset007_FGT_Vessels_0th_1st_sub |
FGT-Vessel-Net-v2 |
Dataset008_FGT_Vessels_0th_1st |
FGT-Vessel-Net-v1 |
Dataset009_FGT_Vessels_ablation |
Model for ablation study (FGT, vessels) |
Dataset100_lesions_ablation |
Model for ablation study (lesions) |
FGT-Vessel-Net-v1 |
Final model for FGT and vessel segmentation |
📝 References
The segmentation framework is based on the original nnU-Net implementation:
Isensee, F., Jaeger, P. F., Kohl, S. A., Petersen, J., & Maier-Hein, K. H. (2021). nnU-Net: a self-configuring method for deep learning-based biomedical image segmentation. Nature Methods, 18(2), 203-211.