基于注意力机制与Swin Transformer模型的腰椎图像分割方法

Lumbar Spine Image Segmentation Method Based on Attention Mechanism and Swin Transformer Model

腰椎图像的精确分割是腰椎间盘疾病自动化诊断的重要前提,现有的分割方法在实际应用于分割任务时仍然存在无法精确分割的问题。对此,本文提出了一种基于注意力机制与Swin Transformer模型的腰椎图像分割网络模型。该模型在卷积网络中引入Swin Transformer模型,使用移动窗口的Transformer模块对卷积提取的高层语义信息进行全局信息建模;然后使用注意力机制对上采样过程中跳过连接中传递的低级特征施加权重,去除背景信息,最终实现腰椎图像的精细分割。实验结果表明,本文的腰椎图像分割方法相似度系数指标达到91.18%,性能优于UNet及其变型网络模型。

Accurate segmentation of lumbar images is an important prerequisite for the automatic diagnosis of lumbar intervertebral disc diseases.Existing segmentation methods still have the problem that they cannot be accurately segmented when they are actually applied to segmentation tasks.Therefore,this paper proposes a lumbar image segmentation network model based on the attention mechanism and the Swin Transformer model.The model introduces the Swin Transformer model into the convolution network,and uses the Transformer module of the moving window to model the global information of the high-level semantic information extracted by convolution;then,the attention mechanism is used to weight the low-level features passed in the skip connection in the up sampling process,remove the background information,and finally realize the fine segmentation of lumbar image.The experimental results show that the similarity index of the lumbar spine image segmentation method in this paper reaches 91.18%,and its performance is better than UNet and its variant network model.