Deep Convolutional Neural Network Based Automatic COVID-19 Detection from Chest X-ray Images

To minimize the rate of death from COVID-19 and stop the disease from spreading early detection is vital. The normal RT-PCR tests for COVID-19 detection take a long time to complete. In contrast to this test, Covid-19 can be quickly detected using various machine-learning technologies. Previous studies only had access to smaller datasets, as COVID-19 data was not readily available back then. Since COVID-19 is a dangerous virus, the model needs to be robust and trustworthy, and the model must be trained on a large and diverse dataset. To overcome that problem, this study combines six publicly available Chest X-ray datasets to produce a larger and more diverse balanced dataset with a total of 68,424 images. In this study, we develop a CNN model that primarily entails two steps: (a) feature extraction and (b) classification, which are used to identify COVID-19 positive cases from X-ray images. The accuracy of this proposed model is 97.58%, which is higher than most state-of-the-art models. © 2022 IEEE.

Medical Diagnosis and Identification of Covid -19 by Intelligent IoT System and Resnet 18 Bilinear Deep Greedy Network

An international health crisis has been caused by the widespread COVID-19 epidemic. COVID-19 patient diagnoses are made using deep learning, although this necessitates a massive radiography data collection in order to efficiently deliver an optimum result. This paper presents a novel Intelligent System with IoT sensors for covid 19 and "Bilinear Resnet 18 Deep Greedy Network,"which is effective with a limited amount of datasets. Despite peculiarities brought on by a small dataset, the suggested approach could successfully combat the anomalies of over fitting and under fitting. The suggested architecture ensures a successful conclusion when the trained model is correctly evaluated using the provided X-ray datasets of COVID-19 cases. The recommended model offers accuracy of 97%, which is superior to existing methodologies. Better precision, recall, and F1 score are provided;which are 98%, 96%, and 96.94% respectively, which is better than other existing methodology. © 2022 IEEE.

Learning from Small Datasets: An Efficient Deep Learning Model for Covid-19 Detection from Chest X-ray Using Dataset Distillation Technique

Ever since the spread of the Coronavirus pandemic popularly known as Covid-19, researchers have dwelled in finding ways to curtail the spread of this disease. The disease has no known treatment but the best way of reducing its spread is by conducting tests, to identify people with positive cases, and isolate them from the general public. Despite the efforts being made by medical practitioners and media houses to provide public awareness, the general public is still shunning away from the covid-19 tests, because of the sentiments rumored about the disease and the complications of the testing process. In some countries, even the cost of the tests is beyond the reach of common citizens or simply not affordable. Researchers proposed cost-effective deep learning models of detecting covid-19 from the chest x-ray images, to serve as a diagnostics aid or an improvised tool in places where the testing materials are not affordable or available. However, the models are very cumbersome, making them expensive to train, the model also suffers from a long inference time. As the matter of diagnosis is critical, it is necessary to provide a new faster models with shorter inference time. Therefore, this paper proposed a novel covid-19 diagnosis using dataset distillation. The model used only 70 instances out of 3,616 available instances in the X-ray dataset, making the model resource inexpensive, and faster to train. The performance of the proposed model achieved 95% accuracy when tested, the model also outperformed the convolutional neural network (CNN) model trained with a full dataset in terms of accuracy. © 2022 IEEE.

Diagnose COVID-19 Based on CT Images Using Transfer Learning

Coronavirus disease is an ongoing pandemic caused by a virus called severe acute respiratory syndrome coronavirus 2. Due to the current global pandemic's perilous state, getting a speedy and precise diagnosis of COVID-19 for everyone who wants to have a COVID-19 test should be the priority. Therefore, building the AlexNet model, which is trained for diagnosing COVID-19 based on CT scans from a large dataset which is composed of 104,009 CT slices coming from 1,489 patients (accuracy is around 67.9%) and a small dataset which is composed of 349 CT images from 216 patients (accuracy is around 62.3 %) would have important implications to help early identification of COVID-19. Moreover, due to the lack of CT scans of positive COVID-19 patients, transferring the learned model parameters from a large dataset to a small dataset contributes to better performance on a small dataset. In our model, the effectiveness of transfer learning is proved by a 1.9% increase in the accuracy of a small dataset. © 2021 IEEE.

Clinical outcome prediction in COVID-19 using self-supervised vision transformer representations

Automated analysis of chest imaging in coronavirus disease (COVID-19) has mostly been performed on smaller datasets leading to overfitting and poor generalizability. Training of deep neural networks on large datasets requires data labels. This is not always available and can be expensive to obtain. Self-supervision is being increasingly used in various medical imaging tasks to leverage large amount of unlabeled data during pretraining. Our proposed approach pretrains a vision transformer to perform two self-supervision tasks - image reconstruction and contrastive learning on a Chest Xray (CXR) dataset. In the process, we generate more robust image embeddings. The reconstruction module models visual semantics within the lung fields by reconstructing the input image through a mechanism which mimics denoising and autoencoding. On the other hand, the constrastive learning module learns the concept of similarity between two texture representations. After pretraining, the vision transformer is used as a feature extractor towards a clinical outcome prediction task on our target dataset. The pretraining multi-kaggle dataset comprises 27499 CXR scans while our target dataset contains 530 images. Specifically, our framework predicts ventilation and mortality outcomes for COVID-19 positive patients using baseline CXR. We compare our method against a baseline approach using pretrained ResNet50 features. Experimental results demonstrate that our proposed approach outperforms the supervised method. © 2022 SPIE.

COVID-19 diagnosis with convolution neural networks using CT images

As COVID-19 has spread worldwide, detecting the patients of COVID-19 and taking effective actions has gained more and more importance. Applying a deep learning framework to detect medical pictures has already been used for years. This paper mainly trained a large number of CT images of patients and normal people on three networks: AlexNet, VGG, and ResNet. Based on PyTorch, we build the network successfully and soon examine the performance of the three networks on the test and validation dataset. Our experiments demonstrate that the ResNet performs the best when detecting the COVID-19 CT images. It reaches the accuracy of 99.5%, which proves that it has a strong fitting ability in our dataset, which is not so large. However, when applying the pre-Trained model from the bigger dataset in a smaller dataset, the accuracy of AlexNet and VGGNet will increase accordingly while the accuracy of ResNet decreases. Though we have made many assumptions about the phenomenon, more experiments are needed after the experiment. © COPYRIGHT SPIE.

Deep Learning Based Model for COVID-19 Pneumonia Prediction with Pulmonary CT Images

COVID-19 pneumonia prediction from computed tomography (CT) images has recently become a crucial part of the computer vision field. Computed tomography images can be used to predict COVID-19 pneumonia as an alternative to traditional testing such as RT-PCR mechanisms that helps to consume time and save more lives as well. The previous researcher faces the overfitting problem while working on a small dataset and deeper convolutional neural networks (CNNs). To overcome this problem, in this work, we consider a large dataset named Corona Hack-Chest X-Ray. And perform a comparative analysis based on fine-tuned CNN models, i.e., VGG16 and miniVGGNet. These models are used based on the transfer learning. For that, the models are initially trained with the ImageNet weights and re-trained with the dataset. The comparative analysis of the models for predicting COVID-19 from CT image is evaluated by the Corona Hack-Chest X-Ray dataset. We trained and test the models with 6368 CT images and evaluate the networks performance in terms of various machine learning metrics. This dataset shows 93% accuracy for the miniVGGNet model while showing 89% accuracy for the VGG16 model. © 2022, The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd.

Data Augmentation and CNN Classification for Automatic COVID-19 Diagnosis from CT-Scan Images on Small Dataset

We present an automatic COVID1-19 diagnosis framework from lung CT images. The focus is on signal processing and classification on small datasets with efforts putting into exploring data preparation and augmentation to improve the generalization capability of the 2D CNN classification models. We propose a unique and effective data augmentation method using multiple Hounsfield Unit (HU) normalization windows. In addition, the original slice image is cropped to exclude background, and a filter is applied to filter out closed-lung images. For the classification network, we choose to use 2D Densenet and Xception with the feature pyramid network (FPN). To further improve the classification accuracy, an ensemble of multiple CNN models and HU windows is used. On the training/validation dataset, we achieve a patient classification accuracy of 93.39%. © 2021 IEEE.

Express Diagnosis of COVID-19 on Cough Audiograms with Machine Learning Algorithms from Scikit-learn library and GMDH Shell tool

The ongoing COVID-19 pandemic and necessity of mass control of population makes to create inexpensive rapid diagnostic methods that could replace or complement existing methods based on clinical studies. In response to this challenge, at the end of 2020 MIT scientists proposed a way to detect COVID-19 sick patients using audio recordings of their cough. They build a binary classifier based on a trained deep neural network that provides 99% precision in detecting sick patients on a dataset of 5000 people (the precision of detecting the healthy ones is not reported). In our study, we propose another technology, which uses: (a) a simple transformation of digital audiograms being matrices 'fequency-time'and (b) typical machine learning algorithms from the popular scikit-learn Python library and the platform GMDH Shell. Objects of consideration are: a large unbalanced dataset (282 sick and 1595 healthy) and a small balanced dataset (174 sick and 193 healthy). In total GMDH-based algorithms demonstrated some better results with both datasets. The winners provides the following precisions of detecting sick/healthy patients [%]: (a) 92/95 on the small dataset and 78/95 on the large data set for the algorithm SVM with a Gaussian kernel;(b) 95/97 on the small dataset and 82/96 on the large data set for the algorithm Random Forest based on GMDH. We suppose these results are promising. © 2021 IEEE.

On the Generation of Medical Dialogs for COVID-19

Under the pandemic of COVID-19, people experiencing COVID19-related symptoms have a pressing need to consult doctors. Because of the shortage of medical professionals, many people cannot receive online consultations timely. To address this problem, we aim to develop a medical dialog system that can provide COVID19-related consultations. We collected two dialog datasets - CovidDialog - (in English and Chinese respectively) containing conversations between doctors and patients about COVID-19. While the largest of their kind, these two datasets are still relatively small compared with generaldomain dialog datasets. Training complex dialog generation models on small datasets bears high risk of overfitting. To alleviate overfitting, we develop a multi-task learning approach, which regularizes the data-deficient dialog generation task with a masked token prediction task. Experiments on the CovidDialog datasets demonstrate the effectiveness of our approach. We perform both human evaluation and automatic evaluation of dialogs generated by our method. Results show that the generated responses are promising in being doctorlike, relevant to conversation history, clinically informative and correct. © 2021 Association for Computational Linguistics.