A Comprehensive Review of Covid-19 Imaging Datasets and Machine Learning Models Publisher



H Aghapanahroudsari HAMED ; M Choubin MORTEZA
Source: Published:2024

Abstract

The COVID-19 pandemic has posed unprecedented challenges for global healthcare systems, necessitating the development of reliable diagnostic tools. Among these, medical imaging, particularly CT scans and radiography, has become indispensable for diagnosing and monitoring COVID-19 patients. This review offers a comprehensive analysis of the most prominent imaging datasets, including COVID-CTset, MosMedData, and COVIDx, which have played a critical role in training machine learning (ML) models for segmentation, classification, and localization tasks. We focus on the application of advanced ML techniques, such as convolutional neural networks (CNNs) for image classification and UNet-based models for segmenting infected lung regions. Additionally, the use of image-guided heatmaps to enhance diagnostic accuracy is examined. The review also addresses challenges associated with dataset variability, imbalanced data, and the current limitations of ML models in clinical practice. Furthermore, most of the available COVID-19 imaging datasets are included in this review, with references, articles, and code repositories provided on GitHub.11https://github.com/Hamed-Aghapanah/COVID-Image-Processing For further research and application. Ultimately, this work aims to offer a thorough overview of the field, emphasizing the diagnostic progression during the pandemic and proposing future directions, such as the use of transformer-based models and explainable AI to improve diagnostic precision and applicability in healthcare settings. © 2025 Elsevier B.V., All rights reserved.