Domain-Invariant Prior Knowledge Guided Attention Networks for Robust Skull Stripping of Developing Macaque Brains
| Autor: | Dinggang Shen, Lufan Liao, Zhenyuan Ning, Tao Zhong, Yu Zhang, Li Wang, Fenqiang Zhao, Gang Li, Yuchen Pei |
|---|---|
| Rok vydání: | 2020 |
| Předmět: |
biology
medicine.diagnostic_test Computer science business.industry Magnetic resonance imaging Pattern recognition Macaque 030218 nuclear medicine & medical imaging 03 medical and health sciences 0302 clinical medicine Neuroimaging biology.animal medicine Skull stripping Segmentation Artificial intelligence Invariant (mathematics) business Distance transform 030217 neurology & neurosurgery |
| Zdroj: | Medical Image Computing and Computer Assisted Intervention – MICCAI 2020 ISBN: 9783030597276 MICCAI (7) |
| DOI: | 10.1007/978-3-030-59728-3_3 |
| Popis: | Non-human primates, especially macaque monkeys, with close phylogenetic relationship to humans, are highly valuable and widely used animal models for human neuroscience studies. In neuroimaging analysis of macaques, brain extraction or skull stripping of magnetic resonance imaging (MRI) is a crucial step for following processing. However, the current skull stripping methods largely focus on human brains, and thus often lead to unsatisfactory results when applying to macaque brains, especially for macaque brains during early development. In fact, the macaque brain during infancy undergoes regionally-heterogeneous dynamic development, leading to poor and age-variable contrasts between different anatomical structures, posing great challenges for accurate skull stripping. In this study, we propose a novel framework to effectively combine intensity information and domain-invariant prior knowledge, which are important guidance information for accurate brain extraction of developing macaques from 0 to 36 months of age. Specifically, we introduce signed distance map (SDM) and center of gravity distance map (CGDM) based on the intermediate segmentation results and fuse their information by Dual Self-Attention Module (DSAM) instead of local convolution. To evaluate the performance, we adopt two large-scale and challenging MRI datasets from rhesus macaques and cynomolgus macaques, respectively, with totally 361 scans from two different scanners with different imaging protocols. We perform cross-validation by using one dataset for training and the other one for testing. Experimental results show the robustness of our plug-and-play method on cross-source MRI datasets without any transfer learning. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|