MOST

Official implementation of MOST: MR reconstruction Optimization for multiple downstream Tasks via continual learning

H. Jeong, SY. Chun, and J. Lee, MOST: MR reconstruction Optimization for multiple downstream Tasks via continual learning, MICCAI 2025, [arXiv]

Dependencies

Use environment.yml for required packages, or create a Conda environment with all dependencies:

conda env create -f environment.yml

Dataset

The processed pickle data can be downloaded from google drive link in https://drive.google.com/drive/folders/1DZlGPPKLTbDy9C2lElwPg7TYzZAIHBn3?usp=drive_link

Pretrained models

The checkpoints can be downloaded from google drive link in https://drive.google.com/drive/folders/1xWOzdJUkliPOtDYBem1EEdnJSph6egDb?usp=drive_link

Usage

MOST downstream task-oriented finetuning

Use train_MOST.sh.

Downstream task evaluation for sequence of finetuning

Use test_MOST.sh.