Name
Abstract | ||
|---|---|---|
Deep neural networks have been widely adopted for automatic organ segmentation from CT-scanned images. However, the segmentation accuracy on some small organs (e.g., the pancreas) is sometimes below satisfaction, arguably because deep networks are easily distracted by the complex and variable background region which occupies a large fraction of the input volume. this paper, we propose a coarse-to-fine approach to deal with this problem. We train two deep neural networks using different regions of the input volume. The first one, the coarse-scaled model, takes the entire volume as its input. It is used for roughly locating the spatial position of the pancreas. The second one, the fine-scaled model, only sees a small input region covering the pancreas, thus eliminating the background noise and providing more accurate segmentation especially around the boundary areas. At the testing stage, we first use the coarse-scaled model to roughly locate the pancreas, then adopt the fine-scaled model to refine the initial segmentation in an iterative manner to obtain increasingly better segmentation. We evaluate our algorithm on the NIH pancreas segmentation dataset with 82 volumes, and outperform the state-of-the-art [18] by more than 4% measured by the Dice-Sorensen Coefficient (DSC). In addition, we report 62.43% DSC for our worst case, significantly better than 34.11% reported in [18]. |
Year | Venue | Field |
|---|---|---|
2016 | arXiv: Computer Vision and Pattern Recognition | Computer vision,Scale-space segmentation,Computer science,Segmentation,Segmentation-based object categorization,Artificial intelligence,Computed tomography,Pancreas |
DocType | Volume | Citations |
Journal | abs/1612.08230 | 1 |
PageRank | References | Authors |
0.34 | 0 | 5 |
Name | Order | Citations | PageRank |
|---|---|---|---|
| Yuyin Zhou | 1 | 1 | 1.02 |
| Ling-Xi Xie | 2 | 429 | 37.79 |
| Wei Shen | 3 | 13 | 8.16 |
| Elliot Fishman | 4 | 1 | 0.34 |
| Alan L. Yuille | 5 | 10339 | 1902.01 |