Enhanced COVID-19 Detection by chest x-ray images using transfer learning-based extracted deep features and information fusion

Document Type

Conference Proceeding

Publication Date

1-1-2023

Publication Title

2023 International Conference on Control, Automation and Diagnosis, ICCAD 2023

Keywords

Chest X-rays (CXRs) images, Computer-Aided Diagnosis (CAD), Coronavirus (COVID-19), Deep features, Majority Voting (MV), Transfer Learning (TL)

Abstract

One of the essential factors to limit the spreading of COVID-19 is an early and accurate diagnosis. Chest X-rays (CXRs) imaging is a common approach to identify COVID19, owing to its ability to detect the respiratory problem as a major symptom of COVID-19 and its public access even in third-world countries. A robust and efficient classification by an intelligent computer-aided model plays a prominent role in facilitating this procedure. In this work, a fusion strategy using Transfer Learning (TL) on a Deep Convolutional Neural Network (DCNN), optimized Ensemble Decision Tree (EDT) and Support Vector Machine (SVM) is introduced to classify the positive and negative COVID-19 cases through using Chest X-rays (CXRs) images. First, a ResNet50 approach is applied to perform a direct classification and to extract deep features. Next, Principal Component Analysis (PCA) is employed on the extracted deep features from the ResNet50 to establish new reduced and uncorrelated feature space. Then, these features are forwarded to SVM and EDT for classification. Hyperparameters of SVM and EDT are optimized by Bayesian Optimization (BO) algorithm. In the last step, Majority Voting (MV) is employed to integrate the classification results and identify COVID19. The main benefit of the proposed COVID19 detection scheme is that the deep features automatically capture COVID19 patterns and improve detection efficiency. In addition, the integrated information from various optimized approaches enhances the classification accuracy and leads to more robust and reliable results.

DOI

10.1109/ICCAD57653.2023.10152327

ISBN

9798350347074

Share

COinS