PSHop: A Lightweight Feed-Forward Method for 3D Prostate Gland Segmentation
Fuente:
arXiv
Guardado en:
| Autores principales: | , , , , , , , , , , |
|---|---|
| Formato: | Preprint |
| Publicado: |
2024
|
| Materias: | |
| Acceso en línea: | |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| _version_ | 1866929286693257216 |
|---|---|
| author | Yang, Yijing Magoulianitis, Vasileios Yang, Jiaxin Xue, Jintang Kaneko, Masatomo Cacciamani, Giovanni Abreu, Andre Duddalwar, Vinay Kuo, C. -C. Jay Gill, Inderbir S. Nikias, Chrysostomos |
| author_facet | Yang, Yijing Magoulianitis, Vasileios Yang, Jiaxin Xue, Jintang Kaneko, Masatomo Cacciamani, Giovanni Abreu, Andre Duddalwar, Vinay Kuo, C. -C. Jay Gill, Inderbir S. Nikias, Chrysostomos |
| contents | Automatic prostate segmentation is an important step in computer-aided diagnosis of prostate cancer and treatment planning. Existing methods of prostate segmentation are based on deep learning models which have a large size and lack of transparency which is essential for physicians. In this paper, a new data-driven 3D prostate segmentation method on MRI is proposed, named PSHop. Different from deep learning based methods, the core methodology of PSHop is a feed-forward encoder-decoder system based on successive subspace learning (SSL). It consists of two modules: 1) encoder: fine to coarse unsupervised representation learning with cascaded VoxelHop units, 2) decoder: coarse to fine segmentation prediction with voxel-wise classification and local refinement. Experiments are conducted on the publicly available ISBI-2013 dataset, as well as on a larger private one. Experimental analysis shows that our proposed PSHop is effective, robust and lightweight in the tasks of prostate gland and zonal segmentation, achieving a Dice Similarity Coefficient (DSC) of 0.873 for the gland segmentation task. PSHop achieves a competitive performance comparatively to other deep learning methods, while keeping the model size and inference complexity an order of magnitude smaller. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2403_15971 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | PSHop: A Lightweight Feed-Forward Method for 3D Prostate Gland Segmentation Yang, Yijing Magoulianitis, Vasileios Yang, Jiaxin Xue, Jintang Kaneko, Masatomo Cacciamani, Giovanni Abreu, Andre Duddalwar, Vinay Kuo, C. -C. Jay Gill, Inderbir S. Nikias, Chrysostomos Image and Video Processing Automatic prostate segmentation is an important step in computer-aided diagnosis of prostate cancer and treatment planning. Existing methods of prostate segmentation are based on deep learning models which have a large size and lack of transparency which is essential for physicians. In this paper, a new data-driven 3D prostate segmentation method on MRI is proposed, named PSHop. Different from deep learning based methods, the core methodology of PSHop is a feed-forward encoder-decoder system based on successive subspace learning (SSL). It consists of two modules: 1) encoder: fine to coarse unsupervised representation learning with cascaded VoxelHop units, 2) decoder: coarse to fine segmentation prediction with voxel-wise classification and local refinement. Experiments are conducted on the publicly available ISBI-2013 dataset, as well as on a larger private one. Experimental analysis shows that our proposed PSHop is effective, robust and lightweight in the tasks of prostate gland and zonal segmentation, achieving a Dice Similarity Coefficient (DSC) of 0.873 for the gland segmentation task. PSHop achieves a competitive performance comparatively to other deep learning methods, while keeping the model size and inference complexity an order of magnitude smaller. |
| title | PSHop: A Lightweight Feed-Forward Method for 3D Prostate Gland Segmentation |
| topic | Image and Video Processing |
| url | https://arxiv.org/abs/2403.15971 |