Optic disc segmentation based on phase-fusion PSPNet

XW Fang, YF Shen, B Zheng, SJ Zhu… - Proceedings of the 2nd …, 2021 - dl.acm.org
XW Fang, YF Shen, B Zheng, SJ Zhu, MN Wu
Proceedings of the 2nd international symposium on artificial intelligence …, 2021dl.acm.org
In the analysis of fundus images, optic disc segmentation is vital to judge eye diseases such
as diabetic retinopathy and glaucoma. Improving the accuracy of optic disc segmentation is
of great significance to the diagnosis of the above diseases. Based on the PSPNet model,
the Phase-Fusion PSPNet network structure is proposed. The network is connected to the
phase upsampling module after the pyramid pooling module, which reduces information
loss and makes the network suitable for segmentation tasks with fuzzy edges. The principle …
In the analysis of fundus images, optic disc segmentation is vital to judge eye diseases such as diabetic retinopathy and glaucoma. Improving the accuracy of optic disc segmentation is of great significance to the diagnosis of the above diseases. Based on the PSPNet model, the Phase-Fusion PSPNet network structure is proposed. The network is connected to the phase upsampling module after the pyramid pooling module, which reduces information loss and makes the network suitable for segmentation tasks with fuzzy edges. The principle of phase upsampling module is to upsample the larger size span step by step and combine it with the corresponding size feature map. iChallenge-PM, iChallenge-AMD, and iChallenge-GON as the training and validation datasets in the paper. The IoU and PA of Phase-fusion PSPNet are 89.93% and 94.94%. Compared with PSPNet, the IoU and PA increased by 1.22% and 1.62% respectively. Experiments show that adding the phase upsampling module makes the model have a better segmentation performance.
ACM Digital Library
顯示最佳搜尋結果。 查看所有結果