G-ResUNet: attention-gate and wavelet-based ResUNet for spinal image segmentation

In order to present the details of images and improve the segmentation performance of spinal CT images, in this paper, a new method based on discrete wavelet transform and deep learning is proposed for image segmentation. The new method is specifically designed for a new model called G-ResUNet, which combines the advantages of ResNet and U-Net and integrates an attention gating mechanism within the residual cells to improve the model’s performance. In addition, for the sake of the problem of blurred edges of CT images, the wavelet transform theory is applied in pre-processing to decompose the images horizontally, vertically, and diagonally for details, thereby enhancing the expression of detailed features so as to reinforce the edge information of the images. Then images processed by wavelet transform is input into the model called G-ResUNet for segmentation. The experimental results show that the model performs well in quality evaluation metrics such as AUC, DICE and IoU compared with the traditional U-Net and U-Net++, and the accuracy reaches 99.36%, which is the optimal outcome among those models and prove the advancement of G-ResUNet. Therefore, the new method not only provides a new approach in the field of medical image segmentation, but also provides new ideas for relevant researches of future.