Real-time Covid-19 diagnosis on embedded IoT platforms

Abstract

COVID-19 continues to pose a persistent global health challenge, where rapid, accurate, and accessible diagnostic tools are crucial for controlling viral transmission. In this study, we present a non-invasive, embedded diagnostic system for COVID-19 detection based on chest CT scan image analysis. The acquired CT images are preprocessed to align with the input dimensions required by four lightweight convolutional neural network (CNN) architectures – ResNet-18, MobileNet, ShuffleNet, and SqueezeNet – selected for their efficiency and suitability in embedded systems. Among these, SqueezeNet achieved a classification accuracy and an f1-score of 99.1%, delivering performance comparable to the other models while offering superior computational efficiency, making it particularly well-suited for real-time, resource-constrained applications. The optimized model was deployed on an NVIDIA Jetson embedded platform to enable on-device, real-time COVID-19 detection at the edge. Diagnostic results are transmitted to the ThingSpeak cloud platform via the MQTT protocol, facilitating continuous, remote health monitoring. Experimental findings confirm the feasibility, accuracy, and real-time operational capability of the proposed embedded system for COVID-19 detection using chest CT scan images.