Abstract:
In fewer than two years, COVID-19, which is widely regarded as the most lethal virus
of the twenty-first century, has been responsible for the deaths of millions of people all
over the world. The novel coronavirus known as SARS-CoV-2 is the causative agent of
the respiratory sickness known as COVID-19. It was first identified in Wuhan, China, in
late December of 2019. According to Hopkins’s projections, the virus will have killed
over one million people by October 2020 and infected about 40,000,000 individuals by
then. This infection has rapidly expanded across China and into other nations since then,
creating a global pandemic in 2020 due to its ease of transmission from person to person
via respiratory droplets. Another contagious lung disease, pneumonia is typically
brought on by a bacterial infection of the alveoli. Pus occurs when infected lung tissue
becomes irritated. Patients typically feel the effects of the virus in their lungs first,
thus chest X-rays can help doctors diagnose the disease. Experts perform physical exams
and use diagnostic tools including chest X-rays, ultrasounds, and lung biopsies to identify
whether or not a patient has these illnesses. In this analysis, we recommend using a
chest X-ray to prioritize people for subsequent RT-PCR testing. It would also aid in the
identification of patients with a high chance of COVID and a false-negative RT-PCR who
require additional testing. It is urgent to create automated technologies that could diagnose
this disease in its early stages, in a non-invasive manner, and in a shorter amount of
time. However, selecting the most accurate models to characterize COVID-19 patients is
challenging due to the inability to compare the outputs of diverse data types and gathering
methods. This is the only way to remedy the issue. As a result, much research has been
conducted to establish an appropriate method for diagnosing and classifying people as
COVID-19-positive, healthy, or affected by other pulmonary lung illnesses. In a few earlier
scholarly works, semiautomatic machine learning techniques with limited precision
were proposed.
In this study, we wanted to develop reliable deep learning approaches, which are a
subset of machine learning and AI that model the way humans acquire knowledge. Data
science encompasses fields like statistics and predictive modeling, two of which benefit
greatly from deep learning. One component of this is what are known as convolutional
neural networks (CNN). Any automatic, reliable, and accurate screening strategy
for COVID-19 infection would be helpful for rapid diagnosis and reducing exposure to
the virus for medical or healthcare personnel. The work takes advantage of a versatile and
successful deep learning approach by employing the CNN model to predict and identify a
patient as being unaffected or impacted by the disease using an image from a chest X-ray.
In order to prove how well the CNN model was trained, the researchers employed a dataset
consisting of 6,000 images with a resolution of 224x224 and 32 batches. Convolutional neural networks (CNNs) were demonstrated to be very effective for medical picture classification.
The authors of this piece propose using convolutional neural networks (CNNs)
to automatically classify chest X-ray images for signs of COVID-19. Using the dataset,
eleven current CNN models—VGG16, VGG19, DenseNet, max poling operation, and
SoftMax activation function—that can distinguish between COVID-19, pneumonia, and
other lung diseases—were first used to identify the symptoms of COVID-19. To avoid
overfitting, we used a stratified 5-fold cross-validation approach, allocating 90 percent of
the dataset to training and 10percent to testing (unseen folds), and validating our model
on 20 percent of the training data. A 95 percent accuracy rate was achieved during performance
training with the trained model. Python’s built-in machine learning functionality
utilizes a confusion matrix. The predictions made by a classification issue are recorded in
a confusion matrix. For each category, the number of correct and incorrect predictions is
represented by a count value. That’s the key to deciphering the matrix of ambiguity. The
confusion matrix illustrates how your classification model generates predictions despite
the uncertainty it faces. The research study can use chest X-ray pictures to identify and
detect COVID-19, normal, and pneumonia infections, according to the results of the tests.
Description:
This thesis submitted in partial fulfillment of the requirements for the degree of Bachelor of Science in Information and Communication Engineering of East West University, Dhaka, Bangladesh