mdcrt/varnet_modl_pytorch

PyTorch implementation of MoDL: Model Based Deep Learning Architecture for Inverse Problems

MoDL

PyTorch implementation of MoDL: Model Based Deep Learning Architecture for Inverse Problems (Not official!)

Official code: https://github.com/hkaggarwal/modl

Reference paper

MoDL: Model Based Deep Learning Architecture for Inverse Problems by H.K. Aggarwal, M.P Mani, and Mathews Jacob in IEEE Transactions on Medical Imaging, 2018

Link: https://arxiv.org/abs/1712.02862

IEEE Xplore: https://ieeexplore.ieee.org/document/8434321/

Dataset

The multi-coil brain dataset used in the original paper is publically available. You can download the dataset from the following link and locate in under the data directory.

Download Link : https://drive.google.com/file/d/1qp-l9kJbRfQU1W5wCjOQZi7I3T6jwA37/view?usp=sharing

Configuration file

The configuration files are in config folder. Every setting is the same as the paper.

Configuration files for K=1 and K=10 are provided. The authors trained the K=1 model first, and then trained the K=10 models using the weights of K=1 model.

Train

You can change the configuration file for training by modifying the train.sh file.

scripts/train.sh

Test

You can change the configuration file for testing by modifying the test.sh file.

scripts/test.sh

Saved models

Saved models are provided.

K=1: workspace/base_modl,k=1/checkpoints/final.epoch0049-score37.3514.pth

K=10: workspace/base_modl,k=10/checkpoints/final.epoch0049-score39.6311.pth