Related Experiment Video
Updated: Jun 22, 2025
11:03
High-resolution In Vivo Manual Segmentation Protocol for Human Hippocampal Subfields Using 3T Magnetic Resonance Imaging
Published on: November 10, 2015
9.3K
DSnet: a new dual-branch network for hippocampus subfield segmentation
Hancan Zhu1,2, Wangang Cheng1, Keli Hu2
1School of Mathematics, Physics and Information, Shaoxing University, 900 ChengNan Rd, Shaoxing, 312000, Zhejiang, China.
Scientific Reports
|July 3, 2024
Summary
A novel deep learning model, DSnet, enhances hippocampal subfield segmentation accuracy. This dual-branch network improves upon existing methods, offering a more precise tool for neurological disorder research and diagnosis.
Area of Science:
- Neuroimaging
- Computational Neuroscience
- Medical Image Analysis
Background:
- The hippocampus is vital for brain function and implicated in neurological disorders.
- Accurate segmentation of hippocampal subfields is crucial for diagnosis and research.
- Existing automatic segmentation tools face challenges with accuracy and time efficiency due to complex structures and similar voxel values.
Purpose of the Study:
- To introduce a novel deep learning-based dual-branch segmentation network (DSnet) for improved hippocampal subfield segmentation.
- To enhance the global perceptual capabilities and long-term dependency modeling in segmentation networks.
- To leverage regional segmentation information to aid subfield segmentation.
Main Methods:
- Developed a dual-branch segmentation network (DSnet) integrating Transformer architecture and a hybrid attention mechanism.
- Employed a dual-branch structure where hippocampal region segmentation informs subfield segmentation.
- Validated the DSnet algorithm on the public Kulaga-Yoskovitz dataset.
Main Results:
- DSnet demonstrated superior performance in hippocampal subfield segmentation compared to conventional single-branch networks.
- The proposed DSnet achieved a 0.57% improvement in average Dice accuracy over the 3D U-Net model.
- Experimental results confirmed the efficacy of the dual-branch approach and integrated attention mechanisms.
Conclusions:
- The DSnet model offers a significant advancement in automated hippocampal subfield segmentation.
- The integration of Transformer and attention mechanisms enhances the network's ability to handle complex neuroanatomical structures.
- DSnet provides a more accurate and efficient tool for clinical diagnosis and neuroscience research involving the hippocampus.
