lungs_segmentation

Automated lung segmentation in chest-x ray

Train pipeline: https://github.com/alimbekovKZ/lungs_segmentation_train

Installation

pip install lungs-segmentation

Example inference

Jupyter notebook with the example:

WebApp

lungs segmentation streamlit

Models weights

model	best dice	Mb
resnet34	0.9657	103.4
densenet121	0.9655	62.8

Usage

Code example for resnet34:

from lungs_segmentation.pre_trained_models import create_model
import lungs_segmentation.inference as inference

model = create_model("resnet34")
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
model = model.to(device)

plt.figure(figsize=(20,40))
plt.subplot(1,1,1)
image, mask = inference.inference(model,'739px-Chest_Xray_PA_3-8-2010.png', 0.2)
plt.imshow(inference.img_with_masks( image, [mask[0], mask[1]], alpha = 0.1))

Results on data from the Internet

resnet34

densenet121

Authors

Renat Alimbekov, Ivan Vassilenko, Abylaikhan Turlassov