Abstract: For the foreseeable future, Covid-19 is likely to be a crucial differential diagnosis for anyone who visits the hospital with symptoms of flu, shortness of breath, conjunctival congestion, fatigue, body ache, lymphopenia on a complete blood count, and/or a change in their typical sense of smell (anosmia) or taste. However, chest radiography of patients who are critically ill and report to the hospital with respiratory symptoms can aid to identify individuals with covid-19 pneumonia. The majority of persons with covid-19 infection do not develop pneumonia and D-dimer was negative. As fast assessment and reporting from an onsite or remote radiologist is not always possible, in this article we provide guidance to non-radiologists on how to look for abnormalities on chest radiographs that may be suggestive of covid-19 pneumonia .Neural Networks (NN) are a subset of Machine Learning that is increasingly being employed in pre-processed image analysis. The CNN (Convolutional Neural Network) algorithm is a common NN technique that outperforms ANN in this project. The existing CNN models are Inception V3, ResNet50, MobileNet, and Xception [1], although they have been proven to be less accurate and time expensive. The H5 model is a new CNN model developed in our Project. A model that was originally created for facial detection and differentiation is currently being utilised to detect all objects with greater accuracy, focusing on five zones with variable pixel intensity scheme. The encouragingly high classification accuracy of our proposal implies that it can efficiently automate Pneumonia Detection in COVID-19 patients from radiograph images to provide a fast and reliable evidence of Pneumonia related COVID-19 infection in the lung that can complement existing COVID-19 diagnostics modalities.
Keywords: H5 Convolutional Neural Network model, Convolution Neural Network (CNN) architecture, COVID-19, Severe Acute Respiratory Syndrome corona virus 2 (SARS cov-2), deep learning based chest radiograph classification (DL-CRC), Tensorflow, Haar Cascade Classifiers, different pixel Intensity scheme, facial detection and distinction.
|
DOI:
10.17148/IJARCCE.2023.12733
