Abstract: The purpose of this research is to develop a novel deep neural network model for identifying COVID disease in patients using chest X-ray images, which remain an affordable and readily available means of diagnosing lung infections. For this purpose, we propose a novel model called CXR-DBICNet (chest X-ray deep binary image classification network). We use EfficentNetV2 as our base layer and build our model on top of that so that it is optimized for binary classification. We use the pre-learned weights and customize them for use with chest X-ray images, which will enable us to extract meaningful features and classify the current problem. We use various layers of convolutions and fully connected layers to achieve the desired model. We evaluated the CXR-DBICNet model using a dataset from Mendeley’s data containing 3271 normal and 1281 COVID images. In the experiment, we achieved a binary accuracy of 96.93%, a precision of 97.50%, a recall of 95.00%, and an F1-score of 96.00%. The results were then compared with five different state-of-the-art EffcientNetV2-based models. The evaluations suggest that the proposed CXR-DBICNet model achieves a superior level of binary classification performance as compared to previous studies, indicating that this model could be effectively used for the clinical preliminary diagnosis of COVID in infected patients.

Keywords: deep learning, convolutional neural networks, pre-learned model, binary classification, EfficientNetV2, chest X-ray

| DOI: 10.17148/IARJSET.2023.107120