Deep Learning-Based Automatic Segmentation of Spinal Magnetic Resonance Images

The accurate segmentation of spinal magnetic resonance (MR) images is a prerequisite for spinal registration, three-dimensional reconstruction, and other technologies. The traditional method of spinal MR image segmentation is cumbersome and has low accuracy. To overcome the drawbacks of traditional methods, a spinal MR image automatic segmentation method based on deep learning is proposed. This method constructs a symmetric channel convolutional neural network to extract multi-scale image features, addresses the network degradation problem during training through residual connections, and reduces information loss by using skip connection layers to connect intermediate features. The network model incorporates a convolutional block attention mechanism to focus on effective features in both spatial and channel dimensions. Experimental results demonstrate that the model achieves an average Dice Similarity Coefficient (DSC) of 0.8619 on the test set, showing an improvement of 15.34%, 7.08%, 5.79%, and 3.1% compared to FCN, U-Net, DeeplabV3+ , and UNet++ network models, respectively. This model can be applied in clinical practice to enhance the segmentation accuracy of spinal MR images.