Automated Brain Extraction for Multi-Contrast MRI in Rat Models Using Enhanced U-Net

Motivation: Brain extraction is an important preprocessing step in preclinical MRI studies. However, there is a lack of reliable tools that perform an accurate brain extraction for small animals, like rats. Goal(s): This study aims to develop a deep-learning model for automated rat brain extraction in multi-contrast MRI scans. Approach: We developed a U-Net model with VGG19 as the encoder. The training was conducted only on T2-weighted images; testing included T2, T1-weighted, and MGE contrasts. Results: The model achieved a dice coefficient of 0.972 on T2 weighted scans. The model also demonstrated high accuracy across different MRI contrasts. Impact: The U-Net model resulted in high segmentation accuracy for various rat brain MRI contrasts. This approach reduced the need for manual brain segmentation for a more automated data processing.