Classification of COVID-19 Disease From Chest X-Ray Images Using the SqueezeNet and Bayesian Optimization Methods

COVID-19 is an infectious disease caused by a viral pathogen. The rapidity of COVID-19's transmission has led to its global pandemic status, affecting numerous countries. Consequently, there is a pressing need for an expeditious and efficacious diagnostic method. According to experts, radiographic images, specifically X-ray images, exhibit ground-glass opacities that can be leveraged for COVID-19 diagnosis. This study entails a dataset categorized into three distinct classes: COVID-19, pneumonia, and normal cases. The classification procedure harnesses Convolutional Neural Networks (CNNs) employing the Squeezenet-Bayesian optimization architecture, herein referred to as COVIDiagnosis-Net. COVIDiagnosis-Net represents a fusion of the SqueezeNet architecture and Bayesian optimization. Bayesian optimization facilitates online training to determine optimal hyperparameters within the SqueezeNet architecture. Once the optimal hyperparameters are established, model training commences to yield the finest model. Our research utilizes a dataset sourced from the Kaggle repository, comprising a total of 4,945 data instances. Subsequently, this dataset is partitioned into three distinct subsets: validation data, training data, and test data, with an 80% training data, 20% test data, and 20% validation data split derived from the training dataset. Upon conducting rigorous experimentation, our model achieved an accuracy rate of 93% and an Fl-Score of 93.5%. Additionally, through evaluation with Receiver Operating Characteristic (ROC) analysis, we obtained an average Area Under Curve (AUC) value of 0.9888 or 98.88%. Interpreting the AUC value, the COVIDiagnosis-Net model can be classified as highly proficient, given its AUC exceeding 0.9(90%).