Patch-based 3D U-Net and transfer learning for longitudinal piglet brain segmentation on MRI.

Area of Science:

• Neuroscience
• Medical Imaging
• Computational Biology

Background:

• Studying neural plasticity in developing brains requires large animal models.
• Piglets, with brains similar to humans, are suitable but underutilized.
• Accurate Magnetic Resonance Imaging (MRI) analysis necessitates automated segmentation algorithms.

Purpose of the Study:

• To develop a fully automatic brain segmentation method for piglets using MRI.
• To enable longitudinal monitoring of brain development in piglets.
• To establish a reliable tool for studying neural plasticity in immature brains.

Main Methods:

• A 3D patch-based U-Net combined with a post-processing pipeline for segmentation.
• Implementation of a transfer-learning strategy for automated longitudinal monitoring across four developmental stages.
• Comparison with 2D U-Net, Brain Extraction Tool (BET), and other animal-specific techniques.

Main Results:

• The proposed 3D U-Net method achieved high accuracy (Dice score: 0.952, Hausdorff distance: 8.51).
• Outperformed 2D U-Net (Dice: 0.919) and BET (Dice: 0.764).
• Transfer learning demonstrated robust performance across developmental stages, particularly in older piglets.

Conclusions:

• A novel method for longitudinal piglet brain segmentation using 3D U-Net and transfer learning was developed.
• This method facilitates future morphometric studies in large animal models.
• The approach is adaptable for segmenting brains of other animal species.