COVID-19 Detection Using Chest X-Ray and Transfer Learning

COVID-19 Detection Using Chest X-Ray and Transfer Learning

Aditya Sharma, Arshdeep Singh Chudey, Mrityunjay Singh
DOI: 10.4018/978-1-7998-3299-7.ch011
OnDemand:
(Individual Chapters)
Available
$37.50
No Current Special Offers
TOTAL SAVINGS: $37.50

Abstract

The novel coronavirus (COVID-19), which started in the Wuhan province of China, prompted a major outbreak that culminated in a worldwide pandemic. Several cases are being recorded across the globe, with deaths being close to 2.5 million. The increased number of cases and the newness of such a pandemic has resulted in the hospitals being under-equipped leading to problems in diagnosis of the disease. From previous studies, radiography has proved to be the fastest testing method. A screening test using the x-ray scan of the chest region has proved to be effective. For this method, a trained radiologist is needed to detect the disease. Automating this process using deep learning models can prove to be effective. Due to the lack of large dataset, pre-trained CNN models are used in this study. Several models have been employed like VGG-16, Resnet-50, InceptionV3, and InceptionResnetV2. Resnet-50 provided the best accuracy of 98.3%. The performance evaluation has been done using metrics like receiver operating curve and confusion matrix.
Chapter Preview
Top

Introduction

Covid-19 originated in Wuhan, China in late 2019 with symptoms ranging from mild illness to pneumonia. The virus can spread through contact, droplets, and fomites. The virus has mild to no symptoms in over 80% of the cases that cause the rapid spreading of the virus (Beluz, J., 2020). This caused the disease to spread in a major part of the population because some carriers didn’t develop any symptoms and hence didn’t take any precautions. The virus has affected over 113 million people worldwide and over 2.5 million deaths have been reported (World Health Organisation, 2021). Symptoms commence 2-14 days after its exposure to the virus. These include chills, fever, sore throat, vomiting, cough diarrhea, etc. This virus mainly affects the respiratory organs and shows its symptoms in the lungs (McKeever, A., 2020). The hospitals are getting filled up at a high rate causing lack of services for the people who are in dire need of medical attention; these people include those who are suffering from fracture, cancer, heart diseases and some other urgent issues. Due to a large number of cases, the hospitals also run short of detection units for Covid-19. Detection of Covid-19 is of utmost importance until the vaccine is available in bulk. This requires a test that is fast and accurate enough to help in curbing the spread. The government of China has approved a reverse transcription-polymerase chain reaction (RT-PCR) which makes use of respiratory and blood samples and applies gene sequencing to it; this procedure has been introduced as the main screening test for the detection of the disease (Ai et al., 2020). The sensitivity of the RT-PCR test is highly variable and findings in China show a poor sensitivity relatively (Huang et al, 2017). In most of the cases, we are not able to detect the COVID-19 that results in a failure to prevent the disease from spreading (Wang et al., 2020). As the proper vaccination is also not available, therefore, it becomes a serious issue for prior detection of the disease and maintaining the protocols.

Covid-19 has symptoms that are similar to other respiratory diseases; therefore, having such symptoms is not a sure sign of having it. Deep learning techniques have made great leaps in the field of image processing, and are helpful to detect complex patterns from the images. These techniques require extensive training with great accuracy. A screening test using the X-Ray scan of the chest region has proved to be effective. For this method, a trained radiologist is needed to detect the patterns which indicate the disease in an X-Ray scan. Automating this process using deep learning models can prove to be effective. It will make the process fast and also help in cases where a trained radiologist isn't present. A screening method consisting of Chest radiography (CT) imaging has been employed in the detection of chest diseases like pneumonia as well. When employed in detection to diagnose of COVID-19 it has a high sensitivity for it (Fang et al., 2020). The Chest CT shows some ground-glass opacities which can be accessed by a trained radiologist (Minaee et al., 2020). This process can be done using a CNN architecture that provides efficient and accurate results. In this study, we use the variant of the existing convolutional neural network (CNN) models that are InceptionV3, InceptionResnetV2, Resnet-50, and VGG-16. This chapter explores the effectiveness of the different CNN models in separating the COVID positive patients from NON-COVID patients. We choose the CNN based models that have high accuracy and effectiveness in the field of image recognition. Due to lack of an appropriate dataset, we pose several problems and try to resolve them by using transfer learning to achieve the high accuracy; this also provides the benefit of less training time. We use the pre-trained CNN models and make use of transfer learning to provide accurate results. We have conducted this study on a publicly available dataset which consists of 142 chest X-ray images of covid positive patients and 142 chest X-ray images of non-covid patients. We have splitted the data set into 80%-20% of training and test sets, then fed it to various pre-trained CNN models to train them. Several models have been employed to diagnose the disease making the process fast and reliable. After the completion of training of the selected models, we have tested our modern on unseen data (i.e., testing data set) and report their accuracy. We achieve the testing accuracy for different CNN models as for VGG -16 96.7%, for Inception-V3 96%, for InceptionResnetV2 76.7% and for Resnet-50 98.3%. The Resnet-50 model has achieved great accuracy of 98.3%.

Complete Chapter List

Search this Book:
Reset