Custom YOLO Object Detection Model for COVID-19 Diagnosis

Abstract

The emergence and spread of the new coronavirus (COVID-19) poses a new public health threat to the entire world (SARS-CoV-2). This new virus is highly contagious and pathogenetically different from other mainstream respiratory viruses. Clinical staff can benefit from Computer Aided Diagnostics (CAD) systems that combine deep learning algorithms and image processing technologies as diagnostic tools for COVID-19. These tools also help to better understand the course of the disease. In most cases, medical staff and healthcare facilities would be more equipped to promptly diagnose COVID-19 for patients with improved flexibility. To examine the training performance of the contemporary YOLOv4 model, this work presents the development of a computer-assisted automatic detection system that focuses specifically on identifying viral cells in blood samples from patients using electron microscopy images to detect the infected blood cell. The mean average precision of the proposed custom model is 86.5%mAP, making it suitable for the upcoming COVID-19 monitoring systems.