COVID-19 detection from CT scans using a two-stage framework

https://doi.org/10.1016/j.eswa.2021.116377Get rights and content

Highlights

  • An end-to-end framework is developed to detect COVID-19 from CT scan images.

  • 3 standard CNN models (DenseNet, ResNet, Xception) are used as feature extractors.

  • Feature selection is done by Harmony Search and Adaptive β-Hill Climbing.

  • The proposed framework is evaluated on the SARS-COV-2 CT-Scan Dataset.

Abstract

Coronavirus disease 2019 (COVID-19) is a contagious disease caused by severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2). It may cause serious ailments in infected individuals and complications may lead to death. X-rays and Computed Tomography (CT) scans can be used for the diagnosis of the disease. In this context, various methods have been proposed for the detection of COVID-19 from radiological images. In this work, we propose an end-to-end framework consisting of deep feature extraction followed by feature selection (FS) for the detection of COVID-19 from CT scan images. For feature extraction, we utilize three deep learning based Convolutional Neural Networks (CNNs). For FS, we use a meta-heuristic optimization algorithm, Harmony Search (HS), combined with a local search method, Adaptive β-Hill Climbing (AβHC) for better performance. We evaluate the proposed approach on the SARS-COV-2 CT-Scan Dataset consisting of 2482 CT scan images and an updated version of the previous dataset containing 2926 CT scan images. For comparison, we use a few state-of-the-art optimization algorithms. The best accuracy scores obtained by the present approach are 97.30% and 98.87% respectively on the said datasets, which are better than many of the algorithms used for comparison. The performances are also at par with some recent works which use the same datasets. The codes for the FS algorithms are available at: https://github.com/khalid0007/Metaheuristic-Algorithms.

Keywords

COVID-19 detection
Convolutional Neural Network
Harmony Search
Adaptive β-Hill Climbing

Cited by (0)

View Abstract