Related Experiment Video
Updated: Dec 13, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
3.2K
Automatic segmentation of brain MRI using a novel patch-wise U-net deep architecture
Bumshik Lee1, Nagaraj Yamanakkanavar1, Jae Young Choi2
1Department of Information and Communications Engineering, Chosun University, Gwangju, Republic of Korea.
Plos One
|August 4, 2020
Summary
A novel patch-wise U-net architecture improves brain magnetic resonance imaging (MRI) segmentation accuracy. This deep learning method enhances local information retention, achieving superior results for brain structure quantification.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Neuroscience
Background:
- Accurate brain magnetic resonance imaging (MRI) segmentation is crucial for quantifying structural changes.
- Deep learning, particularly the U-net architecture, shows significant promise in biomedical image segmentation.
Purpose of the Study:
- To propose a patch-wise U-net architecture for automated brain structure segmentation in structural MRI.
- To enhance local information retention compared to conventional U-net models.
Main Methods:
- Dividing MRI slices into non-overlapping patches.
- Training a U-net model using these patches and corresponding ground truth patches.
- Implementing a non-overlapping patch-wise U-net approach.
Main Results:
- The proposed patch-wise U-net achieved an average Dice Similarity Coefficient (DSC) score of 0.93.
- Outperformed conventional U-net by 3% and SegNet-based methods by 10% on OASIS and IBSR datasets.
Conclusions:
- The patch-wise U-net architecture effectively segments brain structures in MRI.
- This method offers improved accuracy and local information preservation for brain image analysis.
