A robust IoT-based cloud model for COVID-19 prediction using advanced machine learning technique

Abstract

Due to the rapid spread of COVID-19, an urgent need arose for a quick and accurate diagnosis approach. The well-known RT-PCR test is often unavailable and too costly for many suspected cases. In contrast, the chest X-ray scan proves to be a robust and efficient tool for promptly identifying the virus, using various cost-effective tests. This study proposes a hybrid intelligent model for identifying COVID-19 patients based on their chest X-ray scans. The proposed model consists of five phases. Firstly, the X-ray images are preprocessed using a median filter to enhance and remove noise. Secondly, robust invariant and high-level engineering features are extracted using five different feature extraction methods, namely (speeded up robust features (SURF), grey-level co-occurrence matrix (GLCM), histogram of oriented gradients (HoG), segmentation-based fractal texture analysis (SFTA), and local binary pattern (LBP) to help in rapid diagnosis. Then these features are fused to complement the gap of each other and to achieve the best performance. Thirdly, a deep convolutional neural network (DCNN) is used to classify the chest X-ray scans of infected COVID-19 patients. Then, the proposed model is compared with different machine learning techniques under different quality measures. Fourthly, an IoT platform composed of a mobile client application and a backend system is designed. The mobile client application captures the content from the patient side, handles these captured images, and prepares them for rapid sending to the backend side for further processing. The backend system receives the patient images from the client side and applies the proposed model to diagnose the COVID-19 patients. Finally, a cloud-based design for the platform utilizes the modern AWS services characterized by availability, elasticity, coverage, configurability, and reliability. The results show that the proposed model has a good accuracy of around (99.31%) and outperforms the other models. Therefore, the proposed model could be used for real-time automatic early recognition of COVID-19 cases.