Deep Convolutional Neural Networks for COVID-19 Detection from Chest X-Ray Images using ResNetV2

Rakhymzhan, T., Zarrin, Javad, Maktab Dar Oghaz, Mahdi and Babu Saheer, Lakshmi (2022) Deep Convolutional Neural Networks for COVID-19 Detection from Chest X-Ray Images using ResNetV2. In: SAI Computing Conference 2022, Online.

Full text not available from this repository. (Request a copy)
Official URL: https://link.springer.com/chapter/10.1007/978-3-03...

Abstract

COVID-19 has been identified as a highly contagious and rapidly spreading disease around the world. The high infection and mortality rate characterizes this as a very dangerous disease and has been marked as a global pandemic by the world health organization. Existing COVID-19 testing methods, such as RT-PCR are not completely reliable or convenient. Since the virus affects the respiratory tract, manual analysis of chest X-rays could be a more reliable but not convenient or scalable testing technique. Hence, there is an urgent need for a faster, cheaper, and automated way of detecting the presence of the virus by automatically analyzing chest X-ray images using deep learning algorithms. ResNetV2 is one of the pre-trained deep convolutional neural network models that could be explored for this task. This paper aims to utilize the ResNetV2 model for the detection of COVID-19 from chest X-ray images to maximize the performance of this task. This study performs fine-tuning of ResNetV2 networks (specifically, ResNet101V2), which is performed in two main stages: firstly, training model with frozen ResNetV2 base layers, and secondly, unfreezing some layers of the ResNetV2 and retraining with a lower learning rate. Model fine-tuned on ResNet101V2 shows competitive and promising results with 98.50% accuracy and 97.24% sensitivity.

Item Type: Conference or Workshop Item (Paper)
Keywords: Convolutional Neural Networks, Covid19, ResNet, Fine-tuning
Faculty: COVID-19 Research Collection
Faculty of Science & Engineering
Depositing User: Lisa Blanshard
Date Deposited: 13 Jun 2022 09:47
Last Modified: 28 Jul 2022 10:11
URI: https://arro.anglia.ac.uk/id/eprint/707683

Actions (login required)

Edit Item Edit Item