COVID-19 Detection Based on 2D Parallel CNN Model

ABSTRACT

According to Wikipedia Covid-19 Statistics, on May 17, 2021, more than a total of 163 million cases have been confirmed, with more than 3.38 million deaths attributed to Coronavirus. Yet, with such a dreadful number of cases, the world is still facing many challenges, such as scarce medical resources. A machine learning algorithm that utilizes neural networks to diagnose patients' positivity of Covid-19 can be helpful relief of the current medical resource shortage. It is not only time-saving assistance for the frontline physicians to process the diagnosis but also a potential remainder for individuals to pay careful attention to their health status. This paper proposed an application based on a deep two-dimensional parallel convolutional neural network model to classify patients' chest X-ray images. The input dataset consists of two folders Covid19 Negative (∼1600 images) and Covid19 Positive (∼500 images), where each file from the two folders is a chest X-ray radiograph of one individual. The model is established using the Keras API (an Application Programming Interface in the library of Tensorflow) and is optimized to have the least validation loss after the training process. The final experimental results illustrate a validation accuracy of approximately 99% for the model in only 20 epochs. Furthermore, we have created a simple user interface to upload an image and initiate the prediction process. The interface tells the users their probability of catching the Covid, and it also applies Natural Language Processing (NLP) and Speech Synthesis Communication in a chat box for further interactions. © 2021 IEEE.