Analysis and Research of Segmentation Feature Data of FEMORAL CT Image Based on Improved Unet

Accurate extraction of the femur from CT images of lower limb bones is of great significance for 3D reconstruction. To solve the problem that the existing segmentation methods have a poor segmentation effect on femur images with various shapes and complex structures, we propose a new semantic segmentation model, AP-Unet. Based on the Unet network structure, a channel attention mechanism is added to the encoder part to highlight the target features, while suppressing background noise interference, thereby improving the accuracy of deep and shallow information fusion. To retain global context information, a pyramid pooling module is added to the bottom layer of the encoder. A spatial attention mechanism is added to the decoder to determine the positional relationship of features. The mixed loss function is used to solve the problem of serious imbalance of pixel categories caused by large backgrounds. The experimental results show that this method achieves 89.30 and 94.35% in IoU and F1 values, respectively. Compared with FCN-32s, Unet, and Segnet networks, AP-Unet accurately segments images with higher accuracy and better effect.