Name
Title | ||
|---|---|---|
Multi-Path 3d Convolution Neural Network For Automated Geographic Atrophy Segmentation In Sd-Oct Images |
Abstract | ||
|---|---|---|
To automatically segment the geographic atrophy (GA) in spectraldomain optical coherence tomography (SD-OCT) images, we propose a novel segmentation method by designing a multi-path 3D convolution neural network (CNN) model in this paper. Firstly, the 3D patch was fed into the multi-path 3D CNN model as sample to preserve spatial features and overcome the excessive dependence of layer segmentation. Then, an improved classifier was trained by the optimization of network structure and the combination of softmax loss and center loss. The proposed method has been evaluated in two data sets, including fifty-five and fifty-six cubes respectively. For the two data sets, our method obtained the mean overlap ratio (OR) 87.24% +/- 7.95% and 75.89% +/- 15.11%. Compared with the state-of-the-art-algorithms on these two data sets, the mean OR of our results have been improved 5.38% and 5.89% respectively, indicating that our method can get higher segmentation accuracy. |
Year | DOI | Venue |
|---|---|---|
2018 | 10.1007/978-3-319-95933-7_58 | INTELLIGENT COMPUTING THEORIES AND APPLICATION, PT II |
Keywords | Field | DocType |
Image segmentation, SD-OCT, Multi-path 3D CNN, Geographic atrophy | Optical coherence tomography,Data set,Pattern recognition,Softmax function,Convolutional neural network,Segmentation,Computer science,Image segmentation,Artificial intelligence,Classifier (linguistics),Cube | Conference |
Volume | ISSN | Citations |
10955 | 0302-9743 | 0 |
PageRank | References | Authors |
0.34 | 5 | 4 |
Name | Order | Citations | PageRank |
|---|---|---|---|
| Rongbin Xu | 1 | 37 | 10.01 |
| Sijie Niu | 2 | 47 | 10.94 |
| Kun Gao | 3 | 40 | 16.56 |
| Yuehui Chen | 4 | 1167 | 106.13 |