Intelligent diagnosing COVID- 19 Disease Using a Combination of Deep Features and Analyzing Original Component

Document Type : Original Article

Authors

1 Department of Computer, Torbat-e-Heydariyeh branch, Islamic Azad University, Torbat-e-Heydariyeh, Iran

2 Department of Computer Engineering, University of Torbat Heydarieh, Torbat Heydaieh, Iran

3 Faculty of Computer Engineering Shahid Rajaee Teacher Training University,Iran

Abstract

COVID- 19 disease is spreading worldwide and has killed millions of people in a short period of time. Early diagnosis of the disease can accelerate treatment process and prevent the death of patients. The deep learning approaches are used frequently for diagnosing COVID- 19 disease. This study presents a model based on best features extracted from different deep learning approaches. Although deep networks can extract the features of images, but using several learners in parallel will result in detecting the specific features by each learner. In this study, 3 deep convolution networks have been used to extract the features of COVID- 19 disease images. The dimensions of the extracted features have been reduced using the PCA algorithm. Furthermore, classification has conducted by the Perceptron neural network. The study dataset consisted of X-ray images adapted from Github data repository containing 1125 samples in 3 Normal, COVID- 19 and Pneumonia classes. In all networks, 70% of the samples were used for training and 30% for testing. The K-fold method was used to ensure the accuracy of the training process. The results show that the highest accuracy of learners was 96.1% when they are used independently. After combining deep features of learners the accuracy is 97.7%. The use of machine learning-based techniques can lead to the early diagnosis of COVID- 19 disease and contribute to treatment process. This study introduces a combination of deep learning approach to enhance detection accuracy of COVID- 19 disease.

Keywords

Main Subjects

Advanced Signal Processing

Articles in Press, Accepted Manuscript
Available Online from 22 April 2022

Files

Share

How to cite

Statistics