Efficient Multi-Organ Segmentation Using SpatialConfiguration-Net with Low GPU Memory Requirements
Organizations: Gottfried Schatz Research Center: Biophysics, Medical University of Graz, Austria · Institute of Computer Graphics and Vision, Graz University of Technology, Austria
Abstract
Even though many semantic segmentation methods exist that are able to perform well on many medical datasets, they are often not designed for direct use in clinical practice. The two main concerns are generalization to unseen data with a different visual appearance, e.g., images acquired using a different scanner, and efficiency in terms of computation time and required Graphics Processing Unit (GPU) memory. In this work, we employ a multi-organ segmentation model based on the SpatialConfiguration-Net (SCN), which integrates prior knowledge of the spatial configuration among the labelled organs to resolve spurious responses in the network outputs. Furthermore, we modified the architecture of the segmentation model to reduce its memory footprint as much as possible without drastically impacting the quality of the predictions. Lastly, we implemented a minimal inference script for which we optimized both execution time and required GPU memory.
Figures & tables
| Data Split | Center | Phase | # Num. |
| Training ( 361 cases ) | The National Institutes of Health Clinical Center | portal venous phase | 80 |
| Memorial Sloan Kettering Cancer Center | portal venous phase | 281 | |
| Validation ( 50 cases ) | Memorial Sloan Kettering Cancer Center | portal venous phase | 5 |
| University of Minnesota | late arterial phase | 25 | |
| 7 Medical Centers | various phases | 20 | |
| Testing ( 100 cases ) | Memorial Sloan Kettering Cancer Center | portal venous phase | 5 |
| Operating System | Linux Manjaro 5.13.5-1 |
| CPU | AMD Ryzen 9 3900X |
| RAM | 16 4GB |
| GPU | NVIDIA GeForce RTX 2080 Ti |
| CUDA version | 11.4 |
| Programming language | Python 3.9 |
| Deep learning framework | TensorFlow 2.6 |
| Data augmentation methods | Rotation, translation, scaling, elastic deformation, intensity shift and scaling |
| Initialization of the network | He normal initialization |
| Batch size | 1 |
| Patch sampling strategy | full image |
| Patch size | [ ] to [ ] |
| Image spacing | mm |
| Total iterations | 1,000,000 |
| Training | Liver | Kidney | Spleen | Pancreas | ||||
| DSC (%) | NSD (%) | DSC (%) | NSD (%) | DSC (%) | NSD (%) | DSC (%) | NSD (%) | |
| Fold-1 | ||||||||
| Fold-2 | ||||||||
| Fold-3 | ||||||||
| Fold-4 | ||||||||
| Average | ||||||||
| Organ | DSC (%) | NSD (%) |
| Liver | 95.43 4.77 | 80.03 12.24 |
| Kidney | 89.74 13.71 | 77.76 15.89 |
| Spleen | 93.88 14.02 | 86.57 17.55 |
| Pancreas | 75.15 18.52 | 60.00 16.82 |