COVID 19 lung CT images segmentation an‎d identification using deep convolution methods

8/18/2026 12:00:00 AM

As a contagious disease, coronavirus can become a global epidemic. Due to their strong capability in depicting pulmonary infection patterns, Chest CT scans are valuable tools for diagnosing this illness. In CT images, the severe variations in the shape an‎d scale of regions infected by coronavirus, together with their low-contrast boundaries, can reduce the detection accuracy of manual segmentation an‎d even numerous automated techniques. The present research aimed to introduce innovative deep-learning-based architectures that can achieve more accurate boundaries of infected areas by concurrently extracting detailed local features an‎d global dependencies within the image. In this regard, three networks, namely SE-UNETR, SE-HQRSTNet, an‎d dual-path Swin UNETR, were designed. The SE-UNETR model uses transformer blocks an‎d channel attention mechanisms to enhance the model’s concentration on key features an‎d suppress irrelevant features. In the SE-HQRSTNet network, in addition to using the channel attention mechanisms, Swin Transformer blocks an‎d the multi-resolution idea are used to help the model recognize fine details of edges as well as broader structures of infected areas. In the final architecture, which is main philosophy is on utilizing an auxiliary secondary pathway, we investigate the improvement of the modelʹs performance in segmenting infected areas. Ultimately, by combining transformer an‎d Swin Transformer blocks with suitable convolutional structures an‎d providing multi-scale features, SE-UNETR, SE-HQRSTNet, an‎d dual-path Swin UNETR achieved Dice scores of 0.8581, 0.8684, an‎d 0.8723, respectively, in segmenting infection areas. Compared with recent techniques, these outcomes are not only numerically competitive, but also reveal that these frameworks have innovative designs. Among the proposed models, the third model’s achievement demonstrates that providing multi-scale features an‎d multiple paths can play a significant role in increasing the precision an‎d reliability of the model for automated infection detection. Finally, it is expected that the integration of these intelligent diagnostic tools with medical expertise will help physicians to diagnose, monitor, an‎d manage treatment processes with greater speed.

كروناويروس، سي‌تي ريه، تشخيص خودكار، شبكه عصبي كانولوشني، يادگيري عميق

Coronavirus, Lung CT scan, Automatic Detection, Convolutional Neural Network, Deep Learning

Zeinab Momenipour

Dr. Seyed Aliasghar Beheshti Shirazi