SARS-CoV-2, known as coronavirus, causes COVID-19. It is an infectious disease that was first discovered in China in December 2019.1,2,3The World Health Organization (WHO) has also declared it a pandemic.Fig. 1 shows its detailed structure3This new virus quickly spread around the world. Its effects are transmitted to humans through zoonotic flora. The main clinical symptoms of COVID-19 are cough, sore throat, muscle pain, fever and shortness of breath.4,5RT-PCR is commonly used to detect COVID-19. CT and X-rays also play an important role in early and rapid detection of COVID-196However, RT-PCR has a low sensitivity of 60% to 70% and sometimes gives negative results.7,8CT has been observed to be a subtle approach for detecting COVID-19 and may be the best screening tool9.
Artificial intelligence and its subsets play an important role in medicine, and have recently gained prominence due to their use as tools to assist physicians.10,11,12Deep learning techniques have also been used with remarkable results in the detection of many diseases, such as skin cancer detection, breast cancer detection, and lung segmentation.13,14However, due to limited resources and radiologists, staffing each hospital with a clinician is a daunting task. Therefore, automated AI or machine learning methods are needed to mitigate the problem. Eliminating RT-PCR kits also helps reduce waiting times and testing costs. However, thorough preprocessing of CT images is required for best results. Poisson or impulse noise during the acquisition process of these pictures could have severely corrupted the image information.15Recovering this lost information is essential to facilitate post-processing tasks such as object classification and segmentation. Various filtering algorithms have been proposed to deblur and denoise images. The standard median filter (SMF) is one of the most commonly used nonlinear filters.16.
Numerous SMF corrections including Weighted Median and Central Weighted Median (CWM)17,18, has been proposed. The most widely used noise-adaptive soft-switching median (NASM) was proposed.19, achieved optimal results. However, when the noise density exceeds 50%, the quality of the recovered image is significantly degraded. All of these methods are non-adaptive and cannot distinguish between edge pixels, uncorrupted pixels, and corrupted pixels.Recent deep learning ideas presented at20,21,22 Good for recovering images corrupted by fixed-value impulse noise. However, its efficiency decreases with increasing noise density and decreasing Poisson noise normally present in CT images. Furthermore, most of these methods are not adaptive and fail at recovering images degraded by Poisson noise. In the first phase of this work, layer identification by maximum/minimum intensity removal using adaptive filtering windows is proposed, which can process CT images corrupted by high-density impulses and Poisson noise. The proposed method shows good performance both visually and statistically.
Various deep learning techniques are being used to automatically detect COVID-19. A deep learning model employing the COVIDX-Net model, which consists of seven CNN models, was developed to detect COVID-19 in CT scans.This model has high sensitivity and specificity, with an accuracy of 91.7% he can detect COVID-19twenty three. referencetwenty four Demonstrates a deep learning model that obtains 92.4% results in detecting COVID-19. The ResNet50 model istwenty five It also achieved 98% results.Nonetheless, all these tests were slow to diagnose, and information was lost during the acquisition process, resulting in suboptimal results. There are many studies on the detection of26,27,28,29Research published in .30 proposed two different approaches, each using two systems, to diagnose tuberculosis from two datasets. In this study, the PCA) algorithm was first employed to reduce feature dimensionality with the aim of extracting deep features. Then the SVM algorithm was used to classify the features. This hybrid approach achieved an accuracy of 99.2%, a sensitivity of 99.23%, a specificity of 99.41% and her AUC of 99.78%.Similarly, a study published in31 Utilizing various noise reduction techniques, qualitative visual inspection and quantitative parameters such as peak signal-to-noise ratio (PSNR), correlation coefficient (Cr), and system complexity are calculated to compare the results and universal Determines the best denoising algorithm to be applied dynamically. However, these techniques operate on all pixels, whether or not they are contaminated by noise. An automated deep learning approach from computed tomography (CT) scan images for detecting COVID-19 has been proposed.32This method uses an anisotropic diffusion technique to denoise the images and a CNN model to train the dataset. Finally, various models including AlexNet, ResNet50, VGG16, and VGG19 were evaluated in experiments. This method worked well and achieved higher accuracy. However, its performance degrades when the image is contaminated with higher noise densities.33 used four powerful pre-trained CNN models: VGG16, DenseNet121, ResNet50 and ResNet152 for the COVID-19 CT scan binary classification task. In this method, the FastAI ResNet framework was designed to automatically find the optimal architecture using CT images. Additionally, we used transfer learning techniques to overcome long training times. A higher F1 score of 96% was achieved with this method.Deep learning method announced to detect COVID-19 in him using chest X-ray images 34A dataset of 10,040 samples was used in this study. The model has a detection accuracy of 96.43% and a sensitivity of 93.68%. However, its performance degrades dramatically as the density of Poisson noise increases. Convolutional neural network methods used for VGG-19, Inception_V2, and binary classification pneumonia-based transformations of decision tree models are35In this study, X-ray and CT scan image datasets containing 360-degree images were used for COVID-19 detection. Our findings show that VGG-19, Inception_V2, and decision tree models outperform Inception_V2 (78%) and decision tree (60%) models with 91% accuracy.
In this paper, a paradigm for automated COVID-19 screening based on assessment fusion is proposed. The validity and efficiency of all baseline models were improved by the proposed model utilizing majority prediction techniques to eliminate individual model errors. The proposed AFM model requires only chest X-ray images to diagnose he COVID-19 in an accurate and rapid manner.
The rest of this paper is organized as follows. The dataset is described in the Materials and Methods section. The ‘Proposed Method’ section describes the proposed approach, and the ‘Results and Discussion’ section presents empirical results and analysis. The section “Conclusions” describes the conclusions and specific contributions, along with future directions for improving the efficiency of the proposed work.
