Diagnostic of COVID-19 Pneumonia through Convolutional Neural Networks Using Chest X-RAY

Abstract

This Study to emphasizes the need for improved diagnostic protocols and increased awareness to effectively manage COVID-19 and its complications, particularly pneumonia, to alleviate the burden on healthcare systems, underscores the critical importance of early identification of COVID-19 pneumonia as a strategic approach to mitigate devastating impact and fast detection of underlying symptoms. Introducing a novel model for detecting COVID-19 pneumonia, utilizing chest X-ray images available on open-source platform, and convolutional neural networks, enabling precise diagnosis in binary classification settings. Two steps followed to enhance classification accuracy and avoid Overfitting: (1) enlarging the data set while maintaining the balance of the classification scenarios; (2) incorporating regularization techniques and performing hyper-parameter optimization. The model is ideal for deployed locally with limited capacities and without an Internet access. Because of the network size, the model capacity reduced immensely. Comparison to literature, the final model performed better and required a disproportionately higher parameters while reaching a classification accuracy of 99.63% and model sensitivity of 93.75% for the binary cases. The models can be uploaded to a digital platform for quick diagnosis and make up for lack of professionals, and RT-PCR (reverse transcription polymerase chain reaction).