Convid-Net: An Enhanced Convolutional Neural Network Framework for COVID-19 Detection from X-Ray Images

Abstract

This article aims to demonstrate a deep convolutional neural network (CNN) framework namely Convid-Net based on a combination of residual network and parallel convolution (CONV) to detect COVID-19 from chest X-ray images. The proposed architecture can choose optimum features from different parallel CONV and residual connection increasing overall accuracy with less computational expenses. A custom dataset has been created for this work which consists of total 1440 images of COVID-19, 2470 normal images and 2407 chest X-ray images of viral and bacterial pneumonia; collected from different publicly available sources. Augmentation and preprocessing have been applied as well to increase the number of data for better training purposes. Convid-Net has been trained and tested on a prepared augmented dataset which achieved accuracy of 97.99%. The promising result of the proposed system shows that it converges to an overall higher accuracy and can be a very useful method for physicians and radiologists to assist them in rapid detection and diagnosis of COVID-19 from radiography images. These results also indicate that Convid-Net architecture can further be used in other image based classification tasks.