Refining the Accuracy of Chest X-Ray Image Classification Through Layer and Activation Function Optimization

Abstract

Objective: The goal of this article is to provide the Convolutional Neural Network (CNN)-based algorithm carried out to a Chest X-Ray (CXR) dataset to classifies pneumonia. Method: This study explores reduced layers in architecture of Deep Learning Model (DLM). A framework for classifying Chest- X-Ray image dataset with Deep Learning Model (CXR-DLM) proposed to extract features and detect COVID-19. Finding: Deep learning-based models have an exceptional ability to offer an accurate and efficient system for COVID-19 investigations. However, deep learning models affect the classification of small dataset. CXR-DLM solve this problem by designing with all layers and reduced layers learned during the training phase. The 60% of the CXR images carried for training phase and the remaining 40% for testing phase respectively. The testing images achieve an accuracy 99.57%. CXR normal (1341 images) and Covid (3875 images) collected from Kaggle dataset. Novelty: The proposed work CXR-DLM support in the field of radiological imaging of COVID-19 reduces false positive and false negative errors in the detection and diagnosis of this disease. In this work determined exclusive chance to provide rapid, safe diagnostic services to patients then classification using CT images. Keywords: Deep learning; Covid19; CXR images; CXRDLM