Impact of lung segmentation on the diagnosis and explanation of COVID-19 in chest X-ray images

Code for the paper: "Impact of lung segmentation on the diagnosis and explanation of COVID-19 in chest X-ray images"

Data

We did not include the original images used to compose the database evaluated in this work due to its large size. Instead, we provide all references and links directly in the paper.

This repository contains all the scripts developed for the experiments in this paper. The structure is as follows:

• 1_data contains all preprocessing scripts used to standardize the multiple image sources.
• 2_segmentation contains the notebook (code) as well as a cached version of the U-Net Keras model.
• 3_classification contains the notebooks used in the classification experiments in Keras and all the estimated statistical models and plots in R.
• 4_images contains the final segmented images and their respective masks used in our experiments.

Results

Lung segmentation

Database	Jaccard distance	Dice coefficient
Cohen v7labs	0.041 +- 0.027	0.979 +- 0.014
Montgomery	0.019 +- 0.007	0.991 +- 0.003
Shenzhen	0.017 +- 0.008	0.991 +- 0.004
JSRT	0.018 +- 0.011	0.991 +- 0.006
Manually created masks	0.071 +- 0.021	0.964 +- 0.011
Test set	0.035 +- 0.027	0.982 +- 0.014

COVID-19 identification

Class	COVID-19	Lung opacity	Normal	Macro-avg
Segmented - VGG16	0.83	0.88	0.9	0.87
Segmented - ResNet50V2	0.78	0.87	0.91	0.85
Segmented - InceptionV3	0.83	0.89	0.92	0.88
Non-segmented - VGG16	0.94	0.91	0.91	0.92
Non-segmented - ResNet50V2	0.91	0.9	0.92	0.91
Non-segmented - InceptionV3	0.86	0.9	0.91	0.9

XAI

Citation

@article{teixeira2020impact,
  title={Impact of lung segmentation on the diagnosis and explanation of COVID-19 in chest X-ray images},
  author={Teixeira, Lucas O and Pereira, Rodolfo M and Bertolini, Diego and Oliveira, Luiz S and Nanni, Loris and Cavalcanti, George DC and Costa, Yandre MG},
  journal={arXiv preprint arXiv:2009.09780},
  year={2020}
}