Afri-COVID: A Comprehensive COVID-19 Dataset for Statistical Modeling and Public Health Research in Algeria

Abstract

CT imaging is an invaluable tool for screening suspected COVID-19 patients and monitoring individuals undergoing treatment. This approach can potentially spot clinical condition lesions with false-negative RT-PCR test results. Thus, high-quality CT scan data is essential for high-quality lesion detection algorithms. Yet, there is no dataset gathering positive patients from all variants of coronavirus (e.g., Alpha, Beta, Gamma, Delta, Lambda, and Omicron). Moreover, existing data contain many flaws, such as scarcity due to the low number of patients, CT exams and slices, mixed modality (X-ray, CT scan, and MRI), mixed sources and devices, and short acquisition time; which make them unsuitable for machine learning applications. This paper presents Afri-COVID, an Algerian COVID-19 dataset, which represents a large-scale repository containing 2500 anonymized CT images and a roundup of the latest statistical data and trends about COVID-19 world-wide. CT data were acquired between January 1, 2020, and October 31, 2022, from three Algerian cities. Additionally, comparative results on different public CT scan COVID-based datasets are presented to inform the state-of-the-art.