COVID-19 Detection Based on Chest X-Ray Image Classification using Tailored CNN Model

The outbreak of Covid-19 epidemic led to millions
of injuries and deaths and pressure on the health system. The
limited availability of diagnosis tools and expert radiologists
raise the need of using computer aided tools to diagnosis
Covid-19 cases. In this study, a tailored Convolutional Neural
Network (CNN) architecture model is proposed to automatically
detect Covid-19 cases using chest X-Ray (CXR) images. The
proposed CNN model consist of three phases preprocessing,
feature extraction and classification. The proposed CNN model
depends on kernel separability to reduce the training parameters
to a large extent. Furthermore, the proposed model used residual
connection and batch normalization extensively to maintain the
network stability during the training process and provide the
model with the regularization effect in order to reduce the
overfitting. Training process hyperparameter (such as batch size
and learning rate) are determined dynamically. The proposed
architecture is trained using QaTa-Cov19 benchmark dataset
achieving 100% for accuracy, sensitivity, precision and F1-score
with a very low parameter count (150K) compared with the other
methods in the literature.