Modify convolutional neural network model for the diagnosis of multi-classes lung diseases covid-19 and pneumonia based on x-ray images

Abstract EN

COVID-19 is a new virus able to infect both the upper and lower respiratory lobes of ling.

There is a daily increase in cases and deaths in the global epidemic.

A number of the test kits now in use are sluggish and in short supply; hence RT-PCR testing is the most appropriate option.

To avoid a potentially fatal outcome, early detection of COVID-19 is essential.

According to numerous studies, visual markers (abnormalities) on a patient’s Chest X-Ray imaging can be a valuable characteristic of a COVID-19 patient, which can be exploited to discover the virus.

In this research, Convolutional Neural Networks (CNNs) are being proposed to detect the Covid-19 disease based on X-rays images.

The suggested model is based on modified VGG16 architecture for deep feature extraction.

The fine-tuning approach with end-toend training is also 63tilized in the aforementioned deep CNN models.

The suggested model has been trained and evaluated on the dataset contains 7,245 X-ray images, comprising 1,420 Covid-19 cases, 4,167 bacterial cases of Pneumonia, and 1,658 normal cases.

The model is evaluated using metrics such as accuracy, precision, recall, and f1-score.

The proposed model enhanced the accuracy by using less trainable parameters (weights) than the Vgg16 model.

Thus, the time needed for training and testing will be less.

In addition, it achieved a multiclass micro-average of 97% precision, 97% recall, 97% f1-score, and 97% classification accuracy.

The findings obtained show that the proposed strategy outperforms several currently used methods.

This model appears to be convenient and forceful for multiclass classification.