Publication

Generative Unsupervised Anomaly Detection with Coarse-Fine Ensemble for Workload Reduction in 3D Non-contrast Brain CT of Emergency Room

Jongjun Won¹, Jihwan Kim¹, Joonseo Oh¹, Yereen Yo¹, Jieun Yum¹, Joonsang Lee¹, Joon Hyung Park¹, Wooyoung Jo¹, Nam Yoojin¹, Hyunki Lee², Gil-sun Hong², Namkug Kim¹

¹ Department of Convergence Medicine, Asan Medical Center, University of Ulsan College of Medicine, Seoul, Republic of Korea,
² Department of Radiology and Research Institute of Radiology, University of Ulsan Col-lege of Medicine, Asan Medical Center, Seoul, South Korea
International Conference on Medical Image Computing and Computer Assisted Intervention (MICCAI 2025)

Workload Reduction by using 3D Hierarchical DiffusionAutoencoder

This is the codebase for the paper "Generative Unsupervised Anomaly Detection with Coarse-Fine Ensemble for Workload Reduction in 3D Non-contrast Brain CT of Emergency Room".

This repository is composed of model "Coarse-Morphological-Model (CMM)" & "Fine-Grained-Model (FGM)"

Our code parts are mainly in "CMM" and "FGM" illustrated below:

This repository is based on MONAI.

Monai Generative Models

requirements

pip install -r requirements.txt

Train & Test

There are directories for each upstream model and downstream task.

Train:

For the training stage of CMM & FGM models.

CMM

python main.py --batch_size <batch_size> --log_dir <log_dir> -image_size 96

FGM

python main_seg.py --batch_size <batch_size> --log_dir <log_dir> -image_size 256

Test:

For the testing stage of CMM & FGM models.

CMM

python inference.py --session <inference_dataset_categry> -image_size 96

FGM

python inference.py --session <inference_dataset_categry> -image_size 256

Anomaly Cases of CMM

Anomaly Cases of FGM