3 results about "Optic disc segmentation" patented technology

This application discloses a method, apparatus, device, storage medium, and product for optic disc segmentation. The method includes: extracting multi-level features from a fundus image at both local and global levels; generating several levels of local and global features in ascending order; and then fusing the features in descending order to obtain fused features, thereby obtaining the optic disc segmentation result based on the lowest-level fused features. Each level of feature fusion includes: fusing the local and global features of the current level to obtain preliminary fused features; filtering the preliminary fused features of the current level using a spatial attention module; weighting the local and global features of the current level using the filtered data; and then combining the two weighted features with the fused features output from the next higher level to generate the final fused features of the current level. This application can improve the segmentation accuracy of the optic disc.

This invention presents a passive domain adaptive fundus segmentation method based on evidence representation and marginal screening. Under the condition that the source domain training images and their pixel-level annotations are inaccessible, a Student network and a Teacher network are constructed to learn segmentation of the target domain fundus image. The network outputs positive foreground evidence and negative background evidence for each segmentation channel, and constructs a pixel-level Beta distribution. Based on the Beta distribution, foreground probability, evidence margin, and evidence strength are calculated. Reliable positive pseudo-labels are obtained through joint screening, and a weighted segmentation loss is constructed by combining pseudo-label weights. Simultaneously, regularization loss and evidence strength constraint loss are introduced to optimize the Student network. Finally, the exponential moving average coefficient is dynamically adjusted based on the amount of evidence to update the Teacher network parameters. This method can effectively suppress the propagation of pseudo-label noise and improve the accuracy, stability, and cross-domain adaptability of optic cup and optic disc segmentation in the target domain fundus image.

This invention relates to the fields of medical image processing and auxiliary diagnosis of ophthalmic diseases, specifically to a method and system for retinal vessel and optic disc segmentation based on multi-dimensional attention and edge enhancement mechanisms. It first acquires and preprocesses fundus images, then constructs an EIAIU_Net segmentation model integrating a multi-dimensional attention module and an edge enhancement Transformer module. The model is trained based on a paired shuffling consistency strategy, which reduces the dependence on absolute positional information. After inputting the preprocessed image into the model, multi-path features are generated through residual fusion using the SSA-IAI module, and then bidirectional interaction between edge and global features is achieved using the EETF module's dual-branch structure to generate interactive enhancement features. Finally, pixel-level segmentation results are obtained through the output layer. This invention solves the problems of low efficiency and severe loss of detail in traditional methods, and insufficient segmentation accuracy and limited generalization ability of existing deep learning models, enabling automatic segmentation of retinal vessels and optic discs.