Efficient Multi-Organ Segmentation Using SpatialConfiguration-Net with Low GPU Memory Requirements

Even though many semantic segmentation methods exist that are able to perform\nwell on many medical datasets, often, they are not designed for direct use in\nclinical practice. The two main concerns are generalization to unseen data with\na different visual appearance, e.g., images acquired using a different scanner,\nand efficiency in terms of computation time and required Graphics Processing\nUnit (GPU) memory. In this work, we employ a multi-organ segmentation model\nbased on the SpatialConfiguration-Net (SCN), which integrates prior knowledge\nof the spatial configuration among the labelled organs to resolve spurious\nresponses in the network outputs. Furthermore, we modified the architecture of\nthe segmentation model to reduce its memory footprint as much as possible\nwithout drastically impacting the quality of the predictions. Lastly, we\nimplemented a minimal inference script for which we optimized both, execution\ntime and required GPU memory.\n