A comprehensive study on classification of COVID-19 on computed tomography with pretrained convolutional neural networks

Pham, Tuan D.

doi:10.1038/s41598-020-74164-z

Cited by 122 publications

(59 citation statements)

References 21 publications

(29 reference statements)

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…The first presented with 0.2 % relative lung involvement that was contoured in less than ten minutes, and the second presented with 44.9 % relative lung involvement that required three hours to be completed. Further studies creating and validating artificial intelligence systems using neural network algorithms [ [26] , [27] , [28] , [29] ] could facilitate large scale quantification of percentage lung parenchyma compromised.…”

Section: Discussionmentioning

confidence: 99%

Risk of in-hospital death associated with Covid-19 lung consolidations on chest computed tomography – A novel translational approach using a radiation oncology contour software

Sapienza¹,

Nasra²,

Calsavara³

et al. 2021

European Journal of Radiology Open

View full text Add to dashboard Cite

Section: Discussionmentioning

confidence: 99%

Risk of in-hospital death associated with Covid-19 lung consolidations on chest computed tomography – A novel translational approach using a radiation oncology contour software

Sapienza¹,

Nasra²,

Calsavara³

et al. 2021

European Journal of Radiology Open

View full text Add to dashboard Cite

“…In [78], CXR images of COVID-19 of pneumonia and healthy patients are classified using deep neural networks. In [73], cases of COVID-19 infected were predicted with computed tomography(CT) images using AI and CNN methods. In [105], patients with COVID-19 were detected from CXR images using deep CNN.…”

Section: Literature Review Of Mathematical Modelsmentioning

confidence: 99%

The Prediction of the Lifetime of the COVID-19 Virus using Parameter-dependent New Mathematical Models

2021

Preprint

View full text Add to dashboard Cite

This article is about a complex real-world human medical problem that people all over the world face, a major international public Health problem due to the new coronavirus disease 2019(COVID-19), a highly communicable infectious disease between humans. Spreads rapidly among humans of both sexes of all ages, in large masses in the cyclical manner(seasonally) causing disease in susceptible human Hosts affecting most of the organs in humans mainly lungs resulting in Severe Acute Respiratory Syndrome resulting in mass acute deaths. Acute deaths are more common with Comorbidities like Diabetes mellitus, Ischaemic heart disease, Liver disease, Kidney disease, Gut, etc. Now it is the major emergency international pandemic public health medical disease. On the face of the earth, there are large masses of infection and mass acute deaths due to COVID-19 virus infection and so the life of every individual is uncertain at any time. Because of the mass acute deaths from the COVID-19 virus infection, everyone in the world is scared. From now on, it is the responsibility of the researchers of all nations to bring hope to people. In this article, by predicting the lifetime of disease-causing virus, hope to the people is given, to better protect all people and speed up the immediate general pandemic preparedness within the lifespan of the virus. To accelerate actions to save people's lives, mathematical models will help make public health decisions and reduce mortality using the resources available during this time of the COVID-19 pandemic. In this article, to better protect people from disease preparedness for the virus and a general pandemic by predicting the lifetime of the disease-causing coronavirus, three new mathematical models which are dependent on parameters are proposed. The parameters in the model function model uncertainty of death due to the present international real-life problem caused by different strains of the COVID-19 virus. The first model is a model with six parameters and the second and third models are models with seven parameters respectively. These three models are the generalization of the three models of Phem . The errors due to the models of this article are minimized from the errors due to the models of Phem. These three models can predict the acute death count outside the data period and can predict the lifetime. To illustrate the applicability of the models a big data set of size 54 days starting from February 29, 2020, to April 22, 2020, of acute death counts of USA( United States of America) is considered. The main focus is on the USA due to the significant large mass of infection and large mass of acute death from the COVID-19 virus. As a result, everyone's life is uncertain about death at any time. Since it is a major international public health-related medical problem in humans, with an accuracy of 95% of confidence the results using three models are erected. The large mass of acute deaths due to the number of COVID-19 virus infections in the USA are fitted by the model functions of three mathematical models and a solution is found to an international problem. Based on the acute death rate, the lifetime of the COVID-19 virus is estimated to be 1484.76198616309920 days from the first day of acute death, February 29, 2020. In other words, there will be no mass acute deaths from the COVID-19 virus in the USA after April 2024 if the nation follows the guidelines of the WHO(World Health Organization) and the recommendations of the pathogen. And when the people and the government are very well prepared for this crisis then the spread of infection can be prevented, the people and government can be saved from the economic crisis, and many lives can be saved from mass acute deaths. A comparative study of all models is presented for different measures of errors. The acute death count of the USA outside the date of the data set of 54 days is predicted using three models. The data set misses some counts during the collection of data and it is identified. From the ratio of standard deviation and average acute deaths, it is predicted that the total acute death counts during 54 days will be 62,969. Using the standard deviation around the line of regression it is shown that in the data set a large count is missing during the collection of data of USA. Using the coefficient of determination it is predicted that the Model-C, provides 100% of fitness with the given data set and only 0.0% variation. All three models are suitable to fit the data set of acute death counts of the USA, but Model-C is the best and optimal among the three models. Tt is predicted from Model-A, Model-B, and Model-C the total acute death counts during 54 days will be 66537, 67085, and 68523 respectively. Since Model-C is the best and optimal model, the predicted total acute death counts during 54 days will be 68523. Finally, this article suggests various steps to help control the spread and severity of the new disease. The prediction of the lifetime and data count missing in the data set presented in this research article is entirely new and differs totally from all other articles in the literature. To accelerate actions to save people's lives, mathematical models will help make public health decisions and reduce mortality using the resources available during this time of the COVID-19 pandemic.

show abstract

“…Also, Pham in [20] has presented a study of sixteen pre-trained CNNs for COVID-19 classification. In this study, higher classification rates were obtained by using transfer learning instead of data augmentation.…”

Section: Introductionmentioning

confidence: 99%

“… 0.92 GoogLeNet [19] Several pre-trained CNN such as: VGG-19, VGG-16, DenseNet-169, ResNet-50, CTnet-10, InceptionV3 COVID-19: 349 Non-COVID: 216 Acc. 0.945 VGG-19 [20] Pre-trained CNNs such as AlexNet, GoogleNet, ResNet-18, ShuffleNet COVID-19: 347 Non-COVID: 397 Acc. 0.7829 ResNet-18 Prec.…”

Section: Introductionmentioning

confidence: 99%

Efficient deep neural networks for classification of COVID-19 based on CT images: Virtualization via software defined radio

Fouladi

Ebadi

Safaei

et al. 2021

Computer Communications

View full text Add to dashboard Cite

The novel 2019 coronavirus disease (COVID-19) has infected over 141 million people worldwide since April 20, 2021. More than 200 countries around the world have been affected by the coronavirus pandemic. Screening for COVID-19, we use fast and inexpensive images from computed tomography (CT) scans. In this paper, ResNet-50, VGG-16, convolutional neural network (CNN), convolutional auto-encoder neural network (CAENN), and machine learning (ML) methods are proposed for classifying Chest CT Images of COVID-19. The dataset consists of 1252 CT scans that are positive and 1230 CT scans that are negative for COVID-19 virus. The proposed models have priority over the other models that there is no need of pre-trained networks and data augmentation for them. The classification accuracies of ResNet-50, VGG-16, CNN, and CAENN were obtained 92.24%, 94.07%, 93.84%, and 93.04% respectively. Among ML classifiers, the nearest neighbor (NN) had the highest performance with an accuracy of 94%.

show abstract

A comprehensive study on classification of COVID-19 on computed tomography with pretrained convolutional neural networks

Cited by 122 publications

References 21 publications

Risk of in-hospital death associated with Covid-19 lung consolidations on chest computed tomography – A novel translational approach using a radiation oncology contour software

Risk of in-hospital death associated with Covid-19 lung consolidations on chest computed tomography – A novel translational approach using a radiation oncology contour software

The Prediction of the Lifetime of the COVID-19 Virus using Parameter-dependent New Mathematical Models

Efficient deep neural networks for classification of COVID-19 based on CT images: Virtualization via software defined radio

Contact Info

Product

Resources

About