Related Experiment Video
Updated: Oct 26, 2025
06:48
Automated Segmentation of Cortical Grey Matter from T1-Weighted MRI Images
Published on: January 7, 2019
9.1K
Simultaneous brain structure segmentation in magnetic resonance images using deep convolutional neural networks.
Tomoko Maruyama1,2, Norio Hayashi3, Yusuke Sato4,5
1Division of Radiology, Shinshu University Hospital, 3-1-1 Asahi, Matsumoto, Nagano, 390-8621, Japan. tmaruyama@shinshu-u.ac.jp.
Radiological Physics and Technology
|August 2, 2021
Summary
This study demonstrates deep learning for automatic brain organ segmentation in MRI scans. VGG16-weighted SegNet achieved the highest precision, offering an effective approach for evaluating brain changes.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Neuroscience
Background:
- Brain magnetic resonance imaging (MRI) relies on 2D T1-weighted sagittal slices for assessing brainstem atrophy and pituitary gland signals.
- Accurate image segmentation is crucial for automatically tracking chronological changes in the brainstem and pituitary gland.
Purpose of the Study:
- To develop and evaluate deep learning models for automatic segmentation of internal organs in brain MRI.
- Specifically targeting segmentation of the brainstem, corpus callosum, pituitary, cerebrum, and cerebellum in midsagittal 2D T1-weighted images.
Main Methods:
- Employed two deep learning approaches: patch-based segmentation (AlexNet, GoogLeNet, ResNet50) and semantic segmentation (SegNet, VGG16-weighted SegNet, U-Net).
- Evaluated segmentation performance using precision and Jaccard index metrics across six deep convolutional neural network (DCNN) systems.
Main Results:
- VGG16-weighted SegNet achieved the highest precision (0.974), while ResNet50 yielded the lowest (0.506).
- Segmentation using 2D images proved to be a viable and effective method based on calculated metrics and processing times.
- SegNet with VGG16 was identified as the optimal model for automatic organ segmentation in brain sagittal T1-weighted images.
Conclusions:
- Deep learning, particularly VGG16-weighted SegNet, provides an effective solution for automatic segmentation of key brain structures in 2D MRI.
- This automated approach facilitates accurate evaluation of anatomical changes, aiding in neurological assessments.
