A novel Gray-Scale spatial exploitation learning Net for COVID-19 by crawling Internet resources. (March 2022)

• Record Type: Journal Article
• Authors: ElAraby, Mohamed E.; Elzeki, Omar M.; Shams, Mahmoud Y.; Mahmoud, Amena; Salem, Hanaa
• Is Part Of: Biomedical signal processing and control. Volume 73(2022)
• Publication Date: 2022-03
• Subjects: CXR images -- Classification -- COVID-19 -- Web crawler -- Cloud computing -- Deep convolutional neural networks; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2021.103441
• Languages: English
• ISSNs: 1746-8094

Abstract

Graphical abstract: Highlights: Adapting the crawled updated CXR COVID-19 images datasets using web crawler-based cloud environment. Crawling the updated CXR COVID-19 images datasets from different websites simultaneously. Designing a novel Gray-Scale Spatial Exploitation Net (GSEN) to detect infected COVID-19 cases easily. Optimizing the hyperparameters of GSEN by using Stochastic Gradient Descent (SGD) Optimizer. Abstract: Today, the earth planet suffers from the decay of active pandemic COVID-19 which motivates scientists and researchers to detect and diagnose the infected people. Chest X-ray (CXR) image is a common utility tool for detection. Even the CXR suffers from low informative details about COVID-19 patches; the computer vision helps to overcome it through grayscale spatial exploitation analysis. In turn, it is highly recommended to acquire more CXR images to increase the capacity and ability to learn for mining the grayscale spatial exploitation. In this paper, an efficient Gray-scale Spatial Exploitation Net (GSEN) is designed by employing web pages crawling across cloud computing environments. The motivation of this work are i) utilizing a framework methodology for constructing consistent dataset by web crawling to update the dataset continuously per crawling iteration; ii) designing lightweight, fast learning, comparable accuracy, and fine-tuned parameters gray-scale spatial exploitation deep neural net; iii) comprehensive evaluation of the designed gray-scale spatial exploitation net for different collected dataset(s) based on web COVID-19 crawling verse the transfer learning of the pre-trained nets. Different experiments have been performed for benchmarking both the proposed web crawling framework methodology and the designed gray-scale spatial exploitation net. Due to the accuracy metric, the proposed net achieves 95.60% for two-class labels, and 92.67% for three-class labels, respectively compared with the most recent transfer learning Google-Net, VGG-19, Res-Net 50, and Alex-Net approaches. Furthermore, web crawling utilizes the accuracy rates improvement in a positive relationship to the cardinality of crawled CXR dataset.