Multi-organ segmentation over partially labeled datasets with multi-scale feature abstraction
A multi-scale deep learning method for robust multi-organ segmentation when training data is only partially labeled.
A unified multi-scale segmentation framework for learning across multiple partially labeled datasets with a single model.
- Proposed the pyramid-input and pyramid-output feature abstraction network (PIPO-FAN) to reduce semantic gaps across feature scales.
- Introduced a target-adaptive loss with a unified training strategy to enable segmentation over multiple partially labeled datasets using a single model.
- Released the codebase publicly at DIAL-RPI/PIPO-FAN.