Covid-19 X-ray - Two proposed Databases

University of Salento and IEMN DOAE Université Polytechnique Hauts-de-France
Master degree in Computer Engineering
Supervisor: Cosimo Distante, Abdelmalik Taleb-Ahmed
Co-supervisor: Fares Bougourzi, Hadid Abdenour
Student: Edoardo Vantaggiato

---

The recognition of Covid-19 infection from the X-ray images is an emerging field in machine learning and computer vision community. Despite the big efforts that have been made in this field since the appearance of Covid-19 disease (2019), the field still suffers from two drawbacks. First, the available X-ray scans labeled as Covid-19 infected are relatively small. Second, all the works that have been made in the field are separated; no unified data, classes, and evaluation protocol. In this work, based on the public and new collected data, we propose two X-ray covid-19 databases which are: Three-classes Covid-19 ِ and Five-classes Covid-19. For both databases, we test deep learning architectures. In addition, we propose an Ensemble-CNNs approach which outperforms the deep learning architectures and showing promising results in both databases. We make our databases of Covid-19 X-ray scans publicly available to encourage other researchers to use it as a benchmark for their studies.

Our Source

	Source	License
1	ieee8023/covid-chestxray-dataset	Apache 2.0, CC BY-NC-SA 4.0, CC BY 4.0
2	Chest X-Ray Images (Pneumonia) from Kaggle	CC BY 4.0
3	RSNA Pneumonia Detection Challenge from Kaggle	Open Source
4	A Large Chest X-Ray Dataset - CheXpert	Apache 2.0
5	NLM-MontgomerySet	public dataset
6	NLM-ChinaCXRSet	public dataset
7	Algeria Hospital of Tolga	Open Source

Datasets

3-classes dataset

Class	Train original + augmented	Val	Test
Covid-19	404 + 4848	100	207
Normal	404 + 4848	100	207
Pneumonia	404 + 4848	100	207
Total	1212 + 14544	300	621

❗ for class Covid-19, we use as test set unpublished images collected from Hospitals of Tolga, Algeria

5-classes dataset

Class	Train original + augmented	Val	Test
Normal	404 + 4848	100	207
Bacetial Penumonia	404 + 4848	100	207
Viral Pneumonia	404 + 4848	100	207
Covid-19	404 + 4848	100	207
Lung Opacity No Pneumonia	404 + 4848	100	223
Total	2020 + 24240	500	1051

❗ for class Covid-19, we use as test set unpublished images collected from Hospitals of Tolga, Algeria

Contributions

This work was collaboratively conducted by Edoardo Vantaggiato, Emanuela Paladini, Fares Bougourzi, Cosimo Distante, Abdelmalik Taleb-Ahmed, Hadid Abdenour.

Citation

Paper link

@Article{s21051742,
    AUTHOR = {Vantaggiato, Edoardo and Paladini, Emanuela and Bougourzi, Fares and Distante, Cosimo and Hadid, Abdenour and Taleb-Ahmed, Abdelmalik},
    TITLE = {COVID-19 Recognition Using Ensemble-CNNs in Two New Chest X-ray Databases},
    JOURNAL = {Sensors},
    VOLUME = {21},
    YEAR = {2021},
    NUMBER = {5},
    ARTICLE-NUMBER = {1742},
    URL = {https://www.mdpi.com/1424-8220/21/5/1742},
    ISSN = {1424-8220},
    DOI = {10.3390/s21051742}
}