Related Experiment Video
Updated: Aug 5, 2025
Non-invasive Imaging and Analysis of Cerebral Ischemia in Living Rats Using Positron Emission Tomography with 18F-FDG
Published on: December 28, 2014
RU-Net: skull stripping in rat brain MR images after ischemic stroke with rat U-Net
Herng-Hua Chang1, Shin-Joe Yeh2, Ming-Chang Chiang3
1Computational Biomedical Engineering Laboratory (CBEL), Department of Engineering Science and Ocean Engineering, National Taiwan University, No. 1 Sec. 4 Roosevelt Road, Daan, Taipei, 10617, Taiwan. herbertchang@ntu.edu.tw.
A new deep learning tool, Rat U-Net (RU-Net), accurately segments rat brains in MRI scans for stroke research. This automated skull stripping method enhances preclinical stroke analysis by improving pathological brain region extraction.
Area of Science:
- Neuroimaging
- Medical Image Analysis
- Deep Learning
Background:
- Accurate skull stripping of rat brain MRI is crucial for analyzing experimental ischemic stroke models.
- Existing segmentation methods lack reliability, hindering preclinical stroke research.
- A novel, automated tool is needed for efficient extraction of rat brain regions in post-stroke MR images.
Purpose of the Study:
- To develop and evaluate a new deep learning-based skull stripping algorithm, named Rat U-Net (RU-Net), for segmenting rat brains in MRI.
- To provide an efficient and accurate tool for preclinical stroke investigation.
Main Methods:
- The Rat U-Net (RU-Net) employs a U-shaped deep learning architecture integrating batch normalization and residual networks for end-to-end segmentation.
- A pooling index transmission mechanism enhances spatial correlation between the encoder and decoder.
- The algorithm was evaluated on two in-house datasets of diffusion-weighted imaging (DWI) and T2-weighted MRI (T2WI) from 55 subjects each.
Main Results:
- RU-Net demonstrated high segmentation accuracy across diverse rat brain MR images.
- The algorithm achieved superior performance compared to several state-of-the-art methods.
- RU-Net obtained the highest average Dice scores: 98.04% for DWI and 97.67% for T2WI (p < 0.001).
Conclusions:
- The proposed RU-Net is a highly effective tool for automated rat brain skull stripping in MRI.
- This method has the potential to significantly advance preclinical stroke research by enabling efficient pathological brain region extraction.
