kNN-SVM with Deep Features for COVID-19 Pneumonia Detection from Chest X-ray

Most attention has been paid to chest Computed Tomography (CT) in this burgeoning crisis because many cases of COVID-19 demonstrate respiratory illness clinically resembling viral pneumonia which persists in prominent visual signatures on high-resolution CT befitting of viruses that damage lungs. However, CT is very expensive, time-consuming, and inaccessible in remote hospitals. As an important complement, this research proposes a novel kNN-regularized Support Vector Machine (kNN-SVM) algorithm for identifying COVID-induced pneumonia from inexpensive and simple frontal chest X-ray (CXR). To compute the deep features, we used transfer learning on the standard VGG16 model. Then the autoencoder algorithm is used for dimensionality reduction. Finally, a novel kNN-regularized Support Vector Machine algorithm is developed and implemented which can successfully classify the three classes: Normal, Pneumonia, and COVID-19 on a benchmark chest X-ray dataset. kNN-SVM combines the properties of two well-known formalisms: k-Nearest Neighbors (kNN) and Support Vector Machines (SVMs). Our approach extends the total-margin SVM, which considers the distance of all points from the margin;each point is weighted based on its k nearest neighbors. The intuition is that examples that are mostly surrounded by similar neighbors, i.e., of their own class, are given more priority to minimize the influence of drastic outliers and improve generalization and robustness. Thus, our approach combines the local sensitivity of kNN with the global stability of the total-margin SVM. Extensive experimental results demonstrate that the proposed kNN-SVM can detect COVID-19-induced pneumonia from chest X-ray with greater or comparable accuracy relative to human radiologists. © 2022, The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd.

Limitations and implications of doubling time approach in COVID-19 infection spreading study: A gradient smoothing technique

To control the spread of COVID-19, around the world, many countries imposed lockdowns. Numerous studies were reported on COVID-19 in different disciplines with various aspects. The doubling time is a mathematical technique to estimate the current rate of spread of the disease. Researchers used the doubling technique to address the COVID-19 pandemic situation. The larger doubling period represents a low spreading rate, whereas the smaller doubling period represents a high spreading rate. In other words, high infection implies the low doubling period and low infection implies the high doubling period. So, there is an inverse relationship between doubling time and the infection rate. But the real-life data does not follow such a rule properly in various domains. The data shows that after a certain time when the infection is high, the doubling period is also high, which misleads our general concept of doubling time. This chapter addressed this issue by investigating the real-time COVID-19 data. To overcome this limitation, a gradient smoothing technique has been proposed. © 2021, IGI Global. © 2021 by IGI Global.

Neologism Related to COVID-19 Pandemic: A Corpus-Based Study for the Bengali Language

Language is changing over time, and it is a common phenomenon for all languages. Generally, it is a slow but continuous process. The new words come or adopt in the languages, and some existing words become less frequent use or becoming obsolete in written or verbal communication. The term neologism is implying a newly coined word or expression or a phrase that is entering for common use. But sometimes or some special events like War, New disease, Computer, Internet, etc. make the change rapidly, and the COVID-19 pandemic is one such latest event. Note that the infectious disease caused by a newly discovered coronavirus is termed COVID-19, and it is now the official name of coronavirus disease. It has led to an explosion of neologism in the context of disease and several other social contexts. During this period, many new words were coined in the languages and many of these terminologies are rapidly becoming a part of our daily life. For example, some established terms like “lockdown”, “quarantine”, “isolation”, “pandemic”, etc. increased quickly the use in our daily terminology. From the linguistic point of view, the study of such change or adaptation and its quantization is very much important. This study attempted a corpus-based computational approach to explore the adaptation or creation of new words during the outbreak of COVID-19 in the Bengali language. The main components of this work are the creation of the corpus related to the COVID-19 and an algorithm to find out the neologism. For this study, a news corpus has been used. The corpus is created from the news article related to the COVID-19 from January 2020 to February 2021. © 2022, The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd.

Piece-wise linear regression: A new approach to predict COVID-19 spreading

The coronavirus disease 2019 (COVID-19) pandemic is the most rapidly evolving global emergency since March 2020 and one of the most exercised topics in all aspects of the world. So far there are numerous articles that have been published related to COVID-19 in various disciplines of science and social context. Since from the very beginning, researchers have been trying to address some fundamental questions like how long it will sustain when it will reach the peak point of spreading, what will be the population of infections, cure, or death in the future. To address such issues researchers have been used several mathematical models from the very beginning around the world. The goal of such predictions is to take strategic control of the disease. In most of the cases, the predictions have deviated from the real data. In this paper, a mathematical model has been used which is not explored earlier in the COVID-19 predictions. The contribution of the work is to present a variant of the linear regression model is the piecewise linear regression, which performs relatively better compared to the other existing models. In our study, the COVID-19 data set of several states of India has been used. © Content from this work may be used under the terms of the Creative Commons Attribution 3.0 licence. Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI. Published under licence by IOP Publishing Ltd