Automatic brain extraction for rat magnetic resonance imaging data using U²-Net

Area of Science:

• Neuroimaging
• Medical Image Analysis
• Deep Learning

Background:

• Skull stripping is crucial for rodent brain MRI pre-processing.
• Accurate segmentation of intracranial tissue is essential for quantitative analysis.

Purpose of the Study:

• To develop and evaluate a novel deep learning-based skull stripping method for rat brain MRI using U²-Net.
• To compare the performance of the U²-Net method against traditional segmentation techniques.

Main Methods:

• U²-Net model applied to segment 599 rat brain MRI scans.
• Manual labeling of intracranial tissue for training and validation sets (80% train, 20% test).
• Quantitative evaluation using Dice, Jaccard, Sensitivity, Specificity, Pixel Accuracy, and Hausdorff metrics.

Main Results:

• U²-Net demonstrated superior performance compared to RATS and BrainSuite software.
• Achieved high quantitative scores: Dice coefficient 0.9907 ± 0.0016, Jaccard 0.9816 ± 0.0032.
• Excellent specificity (0.9989 ± 0.0002) and low false positive rate (0.0009 ± 0.0002).

Conclusions:

• The U²-Net based method provides a reliable and accurate approach for rat brain MRI skull stripping.
• This deep learning technique enhances the pre-processing pipeline for rodent neuroimaging studies.
• Contributes a valuable tool for researchers analyzing rat brain MRI data.