Abstract
In the battle against COVID-19, chest CT scans play a crucial role in guiding proper treatment and monitoring. However, accurately assessing severity from these scans necessitates the expertise of highly skilled radiologists. This study introduces a groundbreaking deep learning method that utilizes lung CT images to assess COVID-19 severity. This innovative approach presents a robust tool for evaluating lung involvement in COVID-19 patients. Our proposed architecture employs a Bidirectional Long Short-Term Memory Network (BiLSTM) tailored for capturing video information and movement patterns. By harnessing transfer learning from the efficient MobileNetV2 convolutional neural network (CNN) for feature extraction, this method achieves high accuracy in real-time COVID-19 stage detection. Moreover, we introduce a new BiLSTM variant to further enhance the accuracy of stage classification. This variant builds upon established hybrid models recognized for their compactness and effectiveness in extracting relevant features from scans. A substantial dataset of lung CT images, covering various stages of lung involvement across a diverse patient population, was collected over the course of a year during the COVID-19 pandemic. These scans underwent meticulous annotation by expert specialists to determine the percentage of lung involvement, followed by the application of our proposed model to this dataset. This study offers a comprehensive solution for classifying COVID-19 stages and assessing lung involvement. It highlights advancements such as employing MobileNetV2 to capture subtle patterns and deep BiLSTM for recognizing various disease stages. Notably, our hybrid approach achieved a maximum accuracy of 97.41% in distinguishing ten different COVID-19 severity levels. Furthermore, to enhance the efficiency of real-time information processing and performance, edge processing has been incorporated into the proposed model. This addition enables the model to advance in quicker and more accurate identification of disease stages, contributing to the overall enhancement of system performance. The proposed system holds potential as a valuable diagnostic tool for evaluating lung involvement in COVID-19 patients and monitoring disease progression.