MLRU++: Multiscale Lightweight Residual UNETR++ with Attention for Efficient 3D Medical Image Segmentation

Network Design

Results

Synapse Dataset

Qualitative comparison on multi-organ segmentation 3D task.

Qualitative Comparison

BTVC Dataset Results

<hr />

Installation

The code is tested with PyTorch 1.11.0 and CUDA 11.3. After cloning the repository, follow the below steps for installation,

1. Create and activate conda environment

conda create --name mlru_pp python=3.8
conda activate mlru_pp

2. Install PyTorch and torchvision

pip install torch==1.11.0+cu113 torchvision==0.12.0+cu113 --extra-index-url https://download.pytorch.org/whl/cu113

3. Install other dependencies

pip install -r requirements.txt

<hr />

Dataset

We follow the same dataset preprocessing as in UNETR++. We conducted extensive experiments on five benchmarks: Synapse, BTCV, ACDC, and Decathlon-Lung.

Please refer to Setting up the datasets on nnFormer repository for more details.

Training

The following scripts can be used for training our UNETR++ model on the datasets:

bash training_scripts/run_training_synapse.sh
bash training_scripts/run_training_acdc.sh
bash training_scripts/run_training_lung.sh
bash training_scripts/run_training_tumor.sh

<hr />

Evaluation

The checkpoints are avilable here [!(https://drive.google.com/drive/folders/1D_yXZGsHCjAWLHMMnQKAmtKpefv-dzx3?usp=sharing)]

<hr />

Acknowledgement

This repository is built based on nnFormer repository.