CT image registration technology based on spatiotemporal feature fusion for lung detection in martial arts athletes

Pulmonary imaging plays an important role in evaluating the physical condition of martial arts athletes. High precision detection methods can help identify potential minor injuries during training, thereby improving the athletic performance of martial arts athletes and reducing their health risks. A CT image registration method combining spatiotemporal feature fusion was studied and designed to meet the registration requirements of lung images. This method constructs a registration model for lung CT images, uses convolutional long short-term memory networks to model the temporal changes caused by respiratory motion, and embeds a spatiotemporal feature fusion mechanism in the bottleneck layer of the U-Net structure to simultaneously capture spatial details and temporal dynamics. The results indicate that the proposed image registration technique has significant advantages in accuracy, with an average root mean square error of 1.26. The structural similarity index of this research method reaches 0.91, significantly better than the comparative methods. Furthermore, the proposed method achieves RMSEs of approximately 1.12, 1.42, and 2.11 for low-noise, medium-noise, and high-noise image registration, respectively. In contrast, the Multi-Modal Image Registration (MMIR) method achieves RMSEs of 1.74, 1.88, and 2.11 for the three noise levels. The VoxelMorph method achieves RMSEs of 1.47, 1.61, and 1.98 for the same three noise levels. It can be seen that the proposed method exhibits lower errors than the contrasting methods under various noise conditions. Research can help doctors detect hidden lesions in the early stages of diagnosis, reducing the risk of misdiagnosis and missed diagnosis.