The COVID-19 epidemic accelerated the digital change of several services, including healthcare, and increased access to telemedicine. As a result, an increasing number of web tools were introduced to meet patient needs. A safe database can be created in the healthcare industry as a result of digital transformation. This database can be used to protect, store, and share private patient data with healthcare workers, labs, and medical specialists. Designing efficient decision-making tools for COVID-19 diagnostics is now possible thanks to recent developments in information technology and deep learning (DL) models. In this paper, a novel method for diagnosing COVID-19 using deep learning-enhanced eagle perching optimizer (DTH-EPODL) model is presented. With the help of the IoT and the presented DTH-EPODL model, patient information can be gathered and analysed for illness detection. The DTH-EPODL model uses the Gaussian filtering (GF) method to remove noise in the initial step. Additionally, MixNet, a deep convolutional neural network-based method, is used for feature extraction. Using the deep autoencoder (DAE) algorithm, COVID-19 detection and categorization are accomplished. Finally, the DAE approach's associated hyperparameters can be best adjusted using the EPO method, which enhances categorization results. Benchmark chest x-ray datasets can be used to evaluate the experimental validity of the DTH-EPODL method. The experimental results showed that the DTH-EPODL technique outperformed more modern methods.