Transfer Learning-Based Model for Automated COVID-19 Detection Using Computerized Tomography Scan Graph

Abstract

Background: Coronavirus disease 2019 (COVID-19) has had a huge impact on healthcare systems worldwide since 2019. In this study, we discussed how the combination of transfer learning and traditional classifiers performed for image classification of healthy people, COVID-19 patients, pneumonia patients, and lung cancer patients.

Methods: Different combinations of image preprocessing methods and transfer learning architectures were tested and evaluated. The best performed combination was chosen as feature extractor. Features was finally classified by support vector machine (SVM) and optimized by particle swarm optimization (PSO) algorithm.

Results: When combined VGG16 architecture with the PSO-SVM approach, we obtained exciting results, with 93.5% accuracy in recognition.

Conclusions: The experiments’ results suggest VGG16 can reach high accuracy with a small number of epochs. And using VGG16 as a feature extractor then combining it with SVM and appropriate optimization algorithm can improve the classification performance. The new developed classification algorithm may to some extent help clinicians lighten their workload when facing COVID-19 diagnostic problems.

Clin Infect Immun. 2022;7(2):37-48
doi: https://doi.org/10.14740/cii154