High-accuracy and high-precision medical image segmentation is a major research problem in the in-dustry.Unet is an excellent neural network model that has been widely used in medical image segmentation in recent years.In specific applications,practical problems such as small sample size and unbalanced positive and negative samples greatly affect the segmentation accuracy and segmentation precision of the model.Aiming at these problems,an improved Unet medical image segmentation model is proposed.Firstly,the Unet downsampling module is improved,and the hole convolution branch is introduced while applying traditional convolution,which not only expands the receptive field but also preserves spatial features such as position.Secondly,combined with the loss func-tion commonly used in medical image segmentation,channel attention is introduced,so that the model pays more at-tention to the few samples when the positive and negative samples are unbalanced.Finally,experiments are carried out on the finding-lungs-in-ct dataset,and the iou accuracy of the model exceeds 96%,compared with the traditional Unet,the segmentation performance has been greatly improved.
Medical imagingImage segmentationConvolutional neural networkChannel attentionAtrous con-volution
NETL
NSTL
万方数据
