COVID-19 classification using chest X-ray images based on fusion-assisted deep Bayesian optimization and Grad-CAM visualization

Abstract

The COVID-19 virus's rapid global spread has caused millions of illnesses and deaths. As a result, it has disastrous consequences for people's lives, public health, and the global economy. Clinical studies have revealed a link between the severity of COVID-19 cases and the amount of virus present in infected people's lungs. Imaging techniques such as computed tomography (CT) and chest x-rays can detect COVID-19 (CXR). Manual inspection of these images is a difficult process, so computerized techniques are widely used. Deep convolutional neural networks (DCNNs) are a type of machine learning that is frequently used in computer vision applications, particularly in medical imaging, to detect and classify infected regions. These techniques can assist medical personnel in the detection of patients with COVID-19. In this article, a Bayesian optimized DCNN and explainable AI-based framework is proposed for the classification of COVID-19 from the chest X-ray images. The proposed method starts with a multi-filter contrast enhancement technique that increases the visibility of the infected part. Two pre-trained deep models, namely, EfficientNet-B0 and MobileNet-V2, are fine-tuned according to the target classes and then trained by employing Bayesian optimization (BO). Through BO, hyperparameters have been selected instead of static initialization. Features are extracted from the trained model and fused using a slicing-based serial fusion approach. The fused features are classified using machine learning classifiers for the final classification. Moreover, visualization is performed using a Grad-CAM that highlights the infected part in the image. Three publically available COVID-19 datasets are used for the experimental process to obtain improved accuracies of 98.8, 97.9, and 99.4%, respectively.

Multimodal brain tumor detection and classification using deep saliency map and improved dragonfly optimization algorithm

“…Convolutional NN (CNN), a deep learning technique, is made up of several layers, including an input layer, a convolutional layer, a ReLu activation layer, and a fully connected layer. 30 CNN extracts features at both the low and high levels of abstraction. Because CNN models are typically trained on raw images, the possibility of irrelevant feature extraction is extremely high.…”

“…Medical image processing is a hot research topic these days, and researchers have introduced several architectures for various tasks such as enhancement, segmentation, and classification. Convolutional NN (CNN), a deep learning technique, is made up of several layers, including an input layer, a convolutional layer, a ReLu activation layer, and a fully connected layer 30 . CNN extracts features at both the low and high levels of abstraction.…”

Current Technologies for Detection of COVID-19: Biosensors, Artificial Intelligence and Internet of Medical Things (IoMT): Review

“…Screening of SARS-CoV-2 using a lab-bench assay is regarded as the first line of action in terms of minimizing spread and allowing for early treatment of the disease. This prompted the Chinese government to enact several testing points [ 2 , 3 , 4 ].…”

“…Others employ these techniques as a follow-up approach or as confirmation tests. As a result of massive or large-scale screening of radiographic images, these techniques can be tedious for radiologists and can led to misinterpretation [ 3 , 4 , 8 , 9 ].…”

Interactive framework for Covid-19 detection and segmentation with feedback facility for dynamically improved accuracy and trust

“…Chest image processing for Covid-19 detection is focused on identifying any abnormalities in the chest images [ 10 ]. Some common indication of Covid-19 in chest images are—different types of opacity (i.e., subpleural curvilinear opacity, reticulonodular opacity, ground-glass opacity, etc.…”