xmindflow / TransCeption

Official code for Enhancing Medical Image Segmentation with TransCeption: A Multi-Scale Feature Fusion Approach
48 stars 3 forks source link
hierarchical-attention-networks inception medical-image-segmentation segmentation transformer unet-segmentation

Enhancing Medical Image Segmentation with TransCeption: A Multi-Scale Feature Fusion Approach

The official code for "Enhancing Medical Image Segmentation with TransCeption: A Multi-Scale Feature Fusion Approach".

TransCeption is a U-shaped hierarchical architecture which aggregates the inception-like structure in the encoder based on the pure transformer network.

In this approach,

Updates

Citation

@article{azad2023transception,
  title={Enhancing Medical Image Segmentation with TransCeption: A Multi-Scale Feature Fusion Approach},
  author={Azad, Reza and Jia, Yiwei and Aghdam, Ehsan Khodapanah and Cohen-Adad, Julien and Merhof, Dorit},
  journal={arXiv preprint arXiv:2301.10847},
  year={2023}
}

Requirements

This code is implemented in python 3.6.3 using PyTorch library 1.8.0 and tested in ubuntu OS. We use the libraries of these versions:

To set up the correct environment, we recommend running the following code to install the requirements.

pip install -r requirements.txt
cd TransCeption

Dataset preparation

Train

Run the following code to train TransCeption on the Synapse Dataset:

python train_MSTransception.py --dataset Synapse --base_lr 0.05 --max_epochs 500 --eval_interval 20 --model_name TransCeption --batch_size 16 --root_path <your path to ./Synapse/train_npz> --output_dir <your output path>

Test

Run the following code to test the trained TransCeption on the Synapse Dataset:

python test.py --dataset Synapse --base_lr 0.05 --model_name TransCeption --output_dir <your output path> --br_config 2 --weight_pth <your path to .pth file>

We uploaded the trained weight for Synapse dataset here

Quick Overview

Results

Evaluation metrics

Methods

DSC

HD

Aorta

Gallbladder

Kidney(L)

Kidney(R)

Liver

Pancreas

Spleen

Stomach

DARR 69.77 - 74.74 53.77 72.31 73.24 94.08 54.18 89.90 45.96
R50 U-Net 74.68 36.87 87.74 63.66 80.60 78.19 93.74 56.90 85.87 74.16
U-Net 76.85 39.70 89.07 69.72 77.77 68.60 93.43 53.98 86.67 75.58
R50 Att-UNet 75.57 36.97 55.92 63.91 79.20 72.71 93.56 49.37 87.19 74.95
Att-UNet 77.77 36.02 89.55 68.88 77.98 71.11 93.57 58.04 87.30 75.75
R50 ViT 71.29 32.87 73.73 55.13 75.80 72.20 91.51 45.99 81.99 73.95
TransUnet 77.48 31.69 87.23 63.13 81.87 77.02 94.08 55.86 85.08 75.62
SwinUnet 79.13 21.55 85.47 66.53 83.28 79.61 94.29 56.58 90.66 76.60
TransDeepLab 80.16 21.25 86.04 69.16 84.08 79.88 93.53 61.19 89.00 78.40
HiFormer 80.39 14.70 86.21 65.69 85.23 79.77 94.61 59.52 90.99 81.08
MISSFormer 81.96 18.20 86.99 68.65 85.21 82.00 94.41 65.67 91.92 80.81
TransCeption 82.24 20.89 87.60 71.82 86.23 80.29 95.01 65.27 91.68 80.02

Visualization on Synapse

Qualitative result of different models on Synapse dataset. From (a) to (f) are Ground Truth, U-Net, TransUNet, Swin-Unet, MISSFormer, and our TransCeption.

Visualization on ISIC 2018

Visual comparisons of EffFormer baseline and our TransCeption model on the ISIC 2018 Skin Lesion Dataset. Ground truth boundaries are shown in green, and predicted boundaries are shown in blue.

Reference

We thank the authors for the great work of MPViT, Swin-Unet ,and MISSFormer.