Automatic Segmentation of Diabetic Macular Edemas and ILM-RPE Layers in Retinal OCT Images Based on MFANet.

ICCAI(2023)

引用 0|浏览7
暂无评分
摘要
Aiming at the problems of blurred boundary and low segmentation accuracy of existing retinal OCT image segmentation methods, this paper designs a MFANet semantic segmentation model based on encode-decoder for the purpose of accurate segmentation of diabetic macular edemas and ILM-RPE layers, and proposes an automatic retinal OCT image segmentation method based on this model. This method takes the DeepLabV3+ network as the main body and improves the dilated convolution part, which can fully extract the multiscale features of the lesions and obtain a large enough receptive field to effectively balance the local and global information of the retina; the decoder increases the attention feature integration module, reduces information reduction and enlarges the global dimension interaction feature. The introduced global attention effectively strengthens the model's learning of important features of retinal image. The experimental results show that the proposed method has better segmentation effect than the existing methods, and the MIoU reaches 85.82%. It can extract the macular edema regions quickly and accurately, identify the boundary information better, and provide a quantitative analysis tool for clinical diagnosis and treatment.
更多
查看译文
AI 理解论文
溯源树
样例
生成溯源树,研究论文发展脉络
Chat Paper
正在生成论文摘要