A comparative analysis of machine learning algorithms for detecting COVID-19 using lung X-ray images

Hamal, Susmita, Mishra, Bhupesh ORCID: 0000-0003-3430-8989, Baldock, Robert, Sayers, William ORCID: 0000-0003-1677-4409, Adhikari, Tek Narayan and Gibson, Ryan M. (2024) A comparative analysis of machine learning algorithms for detecting COVID-19 using lung X-ray images. Decision Analytics Journal. Art 100460. doi:10.1016/j.dajour.2024.100460 (In Press)

Preview

Text (peer reviewed version) 13943 Hamal, S. et al (2024) A comparative analysis of machine learning algorithms for detecting COVID-19 using lung X-ray images.pdf - Accepted Version Available under License Creative Commons Attribution Non-commercial No Derivatives 4.0. Download (2MB) | Preview

Abstract

Machine intelligence has the potential to play a significant role in diagnosing, managing, and guiding the treatment of disease, which supports the rising demands on healthcare to provide rapid and accurate interpretation of clinical data. The global pandemic caused by the Severe Acute Respiratory Syndrome Coronavirus (SARSCoV-2) exposed a need for rapid clinical data interpretation in response to an unprecedented burden on the healthcare system. A new healthcare challenge has arisen – post-COVID syndrome or ‘long COVID’. Symptoms of the post-COVID syndrome can persist for months following infection with SARS-CoV-2, often characterised by fatigue, breathlessness, dizziness, and pain. Despite this additional healthcare burden, no tests can diagnose, monitor, or determine the efficacy of treatments/interventions to support recovery. In this paper, an array of machine-learning algorithms is trained to evaluate and detect COVID-19-associated changes to lung tissue from X-ray images. X-ray images are classified from open sources into three categories: COVID-19 patients, patients with pneumonia, and unaffected otherwise healthy individuals using existing Machine Learning (ML) and pre-trained deep learning models. Prioritising models with the fewest false positives and false negatives assessed the performance of different models in detecting COVID-19-associated lung tissue. In addition, image pre-processing, data augmentation, and hyperparameter tuning are used to achieve the best accuracy in the models. Different ML models, including K Nearest Neighbour (KNN), and decision trees (DT), as well as transfer learning models such as Convolutional Neural Network (CNN), Visual Geometry Group (VGG-16, VGG-19), ResNet50, DenseNet201, Xception, and InceptionV3, were tested to evaluate the performance of these models for X-ray images classification. The comparative analysis indicates that VGG-19 with augmentation performed best among the ten algorithms with a training accuracy of 99%, testing accuracy of 98%, and precision of 90% for COVID-19, 90% for normal, and 100% for pneumonia. This higher accuracy for detecting COVID-19-associated lung changes on X-ray may be further developed to stratify patients suffering from post-COVID syndrome. This may enable future intervention studies to determine the efficacy of treatments or better track patients’ prognoses to be optimised.

Item Type:	Article
Article Type:	Article
Uncontrolled Keywords:	Machine Learning; Transfer Learning; Convolutional Neural Network; Visual Geometry
Subjects:	Q Science > QA Mathematics > QA75 Electronic computers. Computer science Q Science > QA Mathematics > QA76 Computer software R Medicine > RA Public aspects of medicine
Divisions:	Schools and Research Institutes > School of Business, Computing and Social Sciences
Research Priority Areas:	Applied Business & Technology
Depositing User:	Kamila Niekoraniec
Date Deposited:	23 Apr 2024 14:44
Last Modified:	25 Apr 2024 12:43
URI:	https://eprints.glos.ac.uk/id/eprint/13943

University Staff: Request a correction | Repository Editors: Update this record