Spatial-aware Attention Generative Adversarial Network for Semi-supervised Anomaly Detection in Medical Image

Zerui Zhang, Zhichao Sun, Zelong Liu, Bo Du, Rui Yu, Zhou Zhao, Yongchao Xu

Paper

MICCAI 2024(Early Accept)

Install

1. Clone this repository and navigate to SAGAN folder

git clone https://github.com/zzr728/SAGAN.git
cd SAGAN

2. Install Package: Create conda environment

conda create -n SAGAN python=3.9 -y
conda activate SAGAN
pip install --upgrade pip  # enable PEP 660 support

3. Install additional packages for training cases

pip install -r requirements.txt

Data Preparation

Download the well-processed Med-AD benchmark from: Google Drive | OneDrive.
(RSNA, VinDr-CXR and LAG are one of the benchmarks, and should be only applied for academic research.)

Training

bash train.sh

Evalution

bash test.sh

Pretraining Model Download

This repository provides pre-trained SAGAN model checkpoints that can be downloaded and used for medical anomaly detection tasks.The available pre-trained models include:

Model Descriptions	Model Weights
SAGAN RSNA-finetuned	SAGAN_G_RSNA.ckpt
SAGAN VinDr-CXR-finetuned	SAGAN_G_VinDr-CXR.ckpt
SAGAN LAG-finetuned	SAGAN_G_LAG.ckpt