A Deep Learning based COVID-19 Detection and Classification Framework
Abdul Hannan, Department of Computer Sciences, University of Engineering and Technology, Taxila, Pakistan.
Javed Iqbal, Department of Computer Sciences, University of Engineering and Technology, Taxila, Pakistan.
Farooq Ali, Department of Computer Sciences, University of Engineering and Technology, Taxila, Pakistan.
Iram Abdullah, Department of Computer Sciences, HITEC University, Taxila, Pakistan.
Corresponding Author:
Abdul Hannan (abdulhannan51@hotmail.com)
Abstract:
COVID-19 is a lethal disease that has impacted numerous people around the globe. Due to such severe impact, the World Health Organization (WHO) declared it a pandemic in 2020. Despite the development of vaccines, due to the emergence of new virus strains, the number of fatalities keeps rising. The successful treatment of patients and the overall containment of the pandemic depends on the quick and accurate detection of COVID-19. Since the emergence of the virus, Deep learning (DL) technology has shown to be an important tool for supporting doctors in the diagnosis of viruses and detection of other diseases i.e. pneumonia and chest infections. In this study, we present an automated method for COVID-19 detection from chest X-Rays. The presented approach fine-tunes ResNet-50 architecture on a database containing both normal and virus-infected images. Initially, the images are preprocessed and then augmented to increase the database samples. These images are then supplied to the end-to-end ResNet-50 framework for deep feature extraction and classification. To reach the optimal architecture, we tuned the hyper-parameters by trying different combinations. The system achieved maximum accuracy of 94.1% and performed far better than existing methods in terms of performance. Hence, the suggested COVID-19 detection and classification framework can be deployed in the real-time scenario to reduce the workload of doctors in classifying COVID-19.
Keywords:
Iram COVID19, Deep Learning, ResNet-50, CNN