COVID-19 diagnosis based on CT scan image segmentation using multi-scale convolutional neural network

Abstract

The COVID-19 pandemic continues to pose significant health challenges worldwide. While traditional methods like Reverse Transcription Polymerase Chain Reaction (RT-PCR) are widely used for diagnosis, they often suffer from limitations in speed and accessibility. This chapter proposes an automatic COVID-19 diagnosis technique using CT scan image segmentation, offering a promising alternative for early detection. The proposed method leverages Convolutional Neural Networks (CNNs), enhanced with a multi-scale approach, to capture more detailed semantic features and achieve high-quality segmentation despite challenges such as low image resolution, varying infection characteristics, and limited training data. To address the data scarcity issue, various data augmentation techniques, including scale variation, translation, and rotation, were applied to enhance the model's performance. The proposed CNN was evaluated using the COVID-19 CT dataset, achieving an overall accuracy of 96.85%, with COVID-19 sensitivity at 93.31%, Common Pneumonia sensitivity at 92.93%, and a true negative rate of 97.82%. These results demonstrate the effectiveness and efficiency of the proposed method in detecting COVID-19 at an early stage, reducing the need for human intervention, and accelerating the diagnostic process. Compared to existing methods, this approach offers significant improvements in accuracy and reliability, making it a valuable tool in the ongoing fight against the pandemic.