Diagnosing Pneumonia from Chest X-Rays Using Optimized Residual Network

Chest diseases are thought to be among the most lethal. When we examine the death rate and the enormous number of people who suffer from pneumonia, it becomes clear how critical it is to diagnose the disease early. Recently, academics have started to use deep learning to diagnose medical diseases, while others are working to improve the performance of deep learning neural networks. For many academics and practitioners, optimizing hyperparameters in Convolutional Neural Networks is a time-consuming task. Experts must manually configure a set of hyperparameter options to obtain superior performance hyperparameters. Convolutional Neural Network is used to model and apply the best results of this manual configuration. Various datasets, on the other hand, necessitate different models or a mix of hyperparameters, which can be time-consuming and tiresome. Several models have been developed to handle this, including grid search and random selection. We propose two Residual Networks hyperparameters optimization systems to meet the aims. In order to improve existing diagnosis methods, these optimization techniques are applied to diagnose pneumonia from chest X-rays. To test the method, we employed these algorithms to categorize a COVID-19 and pneumonia dataset made up of X-ray images. The suggested systems demonstrated that adjusting hyperparameters for the ResNet using random search and hyperband optimization algorithms produces better accuracy than other algorithms, with accuracies of 98.84975% and 98.4184%, respectively. We therefore conclude that employing random search or hyperband to optimize ResNet hyperparameters yields better outcomes than other methods.