Research on Lung CT Image Segmentation Technique Based on Improved U-Net and Attention Mechanism

Computed Tomography (CT) is widely used in the diagnosis of lung diseases due to its efficiency and accuracy. However, that is very time-consuming and can hardly satisfy the needs of modern healthcare manually. Therefore, the focus is on developing an automatic infected region segmentation system for the lung. This paper presents an improved version of the U-Net structure by introducing coordinate attention mechanisms to make the image segmentation of the Lung Infection Region Segmentation System more accurate with reduced mistakes. Based on the Kaggle COVID-19 image dataset, this experiment will evaluate the standard U-Net and an improved model for their segmentation effect in small targets with complex backgrounds. The experiment results showed that the improvement in metrics such as mean IoU and mean Pixel Accuracy was small, particularly for small lesions and regions with noise. This chapter has indicated that the attention mechanism is surely enhancing focus on the most important area and hence would provide new optimization for image analysis of medical evidence.