MI2RLNet

This MI2RLNet is the hub of pretrained models in the medical domain.

We hope MI2RLNet helps your downstream task.

• Organizing Team : MI2RL, Asan Medical Center(AMC), Seoul, Republic of Korea

• Contributor

  • Commiter : Kyuri Kim, Jiyeon Seo, Jooyoung Park, Mingyu Kim, Kyungjin Cho, Daeun Kim, Yujin Nam.

  • Reviewer : Sungman Cho, Sungchul Kim.

  • Data Maintainer : Miso Jang, Namkug Kim.

    
    

• Docker images

  • Dockerfille : tensorflow > 2.x, Pytorch 1.x

    
    

Contents

Data description

Modality	Part	Module	Data Reference
X-ray	Chest	L/R Mark Detection	AMC
X-ray	Chest	PA / Lateral /Others Classification	AMC
CT	Chest	Enhanced / Non-Enhanced Classification	AMC
CT	Chest	Lung Segmentation	AMC
CT	Abdomen	Kidnet & Tmuor Segmentation	KiTS 2019
CT	Abdomen	Liver Segmentation	AMC, LiTS 2017
Endoscopy	Abdomen	Polyp Detection	Kvsair-SEG
MR	Brain	Brain Extraction	AMC
MR	Brain	Blackblood Segmentation	AMC

Experiment results

Modality	Part	Module	Results	Wiki	Weights	Framework
X-ray	Chest	L/R Mark Detection	0.99 (mAP)	link	link	TF 2.x
X-ray	Chest	PA / Lateral / Others Classification	0.94 (Acc, external)	link	link	TF 2.x
CT	Chest	Enhanced / Non-Enhanced Classification	0.96 (Acc, external)	link	link	TF 2.x
CT	Chest	Lung Segmentation	0.98 (DSC)	-	link	TF 2.x
CT	Abdomen	Kidney & Tumor Segmentation	0.83 (DSC)	link	link	TF 2.x
CT	Abdomen	Liver Segmentation	0.97 (DSC)	link	link	TF 2.x
Endoscopy	Abdomen	Polyp Detection	0.70 (DSC)	link	link	Pytorch
MR	Brain	MRI/MRA BET (Brain Extration Tool)	0.95 (DSC)	link	MRI MRA	Pytorch
MR	Brain	Blackblood Segmentation	0.83 (DSC)	link	link	TF 2.x

How can we use ?

• The example code below applies to almost all modules. Some modules may require additional parameters.

Inference

from medimodule.Abdomen import LiverSegmentation

# Initialize the model.
# If pre-trained weight exists, enter it together when the model is assigned.
model = LiverSegmentation("/path/of/weight")

# Get a result.
# If you want to save the result, enter it with `save_path` kwargs.
image, mask = model.predict("/path/of/image", save_path="/path/for/save")

Transfer Learning

# Import any module you want to fine-tune.
from medimodule.Abdomen import LiverSegmentation

# Initialize the model with pre-trained weight.
model = LiverSegmentation("/path/of/weight")

# Construct your custom training code.
...
model.train()
...

Contributing

If you'd like to contribute, or have any suggestions for these guidelines, you can contact us at [email protected] or open an issue on this GitHub repository.

All contributions welcome! All content in this repository is licensed under the Apache 2.0 license.