Grey Wolf Optimization Based Hyper-Parameter Optimized Deep EfficientNet for Chest X-Ray Based Detection of COVID-19

The global economy and public health have been severely affected by the COVID-19 epidemic. Inspired by the free resource community's work assembling the COVID-19 data sets and the accomplishments of deep learning methods in successful classification and detection of diseases on medical domain, we embarked on a mission to develop a COVID-19 detection system from Chest X-ray images that could revolutionize the diagnosis and treatment of infectious diseases. We created an enhanced deep EfficientNet-B0 based Convolutional Neural Network (CNN) architecture to detect COVID-19 from chest X-ray pictures that facilitate accurate diagnosis of COVID-19. A compound coefficient is used by the CNN design and development method EfficientNet-B0 to precisely measure each depth, breadth, and resolution component. Using a set of predefined scaling coefficients, its scaling technique uniformly modifies the network's breadth, depth, and resolution. However, the hyperparameters of EfficientNet-B0 must be fine-tuned to optimize its performance. An automated hyperparameter optimized EfficientNet-B0 is proposed to classify COVID-19 from several public datasets of chest X-ray images where the hyperparameters are optimized using a Grey Wolf Optimizer (GWO). When tested on various chest X-ray image datasets, the GWO-EfficientNet-B0 model outperformed the Particle Swarm Optimization (PSO-EfficientNet-B0), and Genetic Algorithm (GA-EfficientNet-B0) required to optimize EfficientNet-B0's hyperparameters in terms of difference performance metrics. The proposed optimized neural network can accurately classify COVID-19 positive cases with 97.89% accuracy rate which is better compared to other methods.