CT image segmentation based on modified U-Net for novel coronavirus pneumonia

Computed Tomography (CT) provides high-resolution pulmonary imaging that aids in identifying novel coronavirus pneumonia patient lung infections, assessing disease severity, and monitoring treatment effects. However, traditional image segmentation methods struggle with complex lesion structures, and while deep learning methods perform well, they often rely on large amounts of annotated data. This study proposes an improved U-Net model for the segmentation of novel coronavirus pneumonia CT images. The proposed model enhances segmentation accuracy by maintaining consistent resolution between input and output images, adding convolutional layers and skip connections, and incorporating a composite loss function. Experimental results show that the improved U-Net model achieves a validation accuracy of 91% and a validation IoU of 84% in the task of novel coronavirus pneumonia CT image segmentation. This model aids in early detection and precise localization of lung lesions, thus improving diagnostic accuracy and timeliness. Additionally, it holds potential for application in the image segmentation tasks of other diseases.