In this research, we have proposed the role of bio-inspired algorithms in deep learning and finalized the hyperparameters during the training process. The selection of hyperparameter values is a tedious task. The recent studies highlighted the role of bio-inspired algorithms in selecting hyperparameter values. The algorithms utilized in this research are grey wolf optimization (GWO) and cuckoo search algorithm (CSA) and a comparative analysis is also done to get the best model out of them. COVID-19 and viral pneumonia patients' chest X-ray (CXR) images exhibit a similar pattern, which often leads to misdiagnosis. The experiments are performed on a pre-trained CNN model. The pre-processing, like histogram equalization and image resizing techniques, is performed over 2860 lung CXR images to classify them into three different classes, namely, COVID-19, pneumonia, and normal images.