Study on COVID-19 vs Viral PNEUMONIA Detection Using Convolutional Neural Network

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.