An Effective ResNet Model for Respiratory Disease Detection: A Case Study on COVID-19 Chest X-ray Images

Bushra Hassan; Kaveh Kiani; Taha Mansouri

doi:10.47852/bonviewAIA52024885

Authors

Bushra Hassan School of Science, Engineering and Environment, University of Salford, UK
Kaveh Kiani School of Science, Engineering and Environment, University of Salford and Data Science and AI HUB, University of Salford, UK
Taha Mansouri School of Science, Engineering and Environment, University of Salford and Data Science and AI HUB, University of Salford, UK

DOI:

https://doi.org/10.47852/bonviewAIA52024885

Keywords:

AlexNet, VGG, ResNet, transfer learning, convolutional neural network, computer vision, computer-aided diagnosis, respiratory disease detection

Abstract

The COVID-19 pandemic has placed an extraordinary burden on healthcare systems worldwide and underscored the urgent need for faster, more reliable diagnostic methods for respiratory diseases. In this study, we explored how artificial intelligence, particularly deep convolutional neural networks (CNNs), can assist in diagnosing COVID-19 using chest X-ray images. Although new CNN architectures continue to emerge, residual neural network (ResNet) models (ResNet50, ResNet101, and ResNet152) remain popular because of their balance of accuracy and robustness. We compared their performance with other established models such as VGG (VGG11 and VGG16) and AlexNet to examine the trade-offs between architectural complexity and practical effectiveness. Among these, ResNet101 achieved the best results, with an accuracy of 93%, a precision of 91.27%, and an AUC–ROC of 97.41%, making it the leading model in our study. Despite promising outcomes, we also highlighted current limitations in deep learning for medical imaging, such as small sample sizes and limited clinical validation. Our findings aim to support the future development of reliable AI-assisted tools in medical diagnostics and highlight the need for more comprehensive validation and ethical considerations.

Received: 22 November 2024 | Revised: 11 April 2025 | Accepted: 5 June 2025

Conflicts of Interest

The authors declare that they have no conflicts of interest to this work.

Data Availability Statement

The data that support the findings of this study are openly available in Kaggle at https://www.kaggle.com/datasets/donjon00/covid19-detection?select=Data.

Author Contribution Statement

Bushra Hassan: Conceptualization, Methodology, Software, Validation, Formal analysis, Investigation, Data curation, Writing – original draft, Visualization. Kaveh Kiani: Conceptualization, Methodology, Investigation, Resources, Writing – review & editing, Visualization, Supervision, Project administration. Taha Mansouri: Conceptualization, Methodology, Investigation, Resources, Supervision.