U-Net based vessel segmentation for murine brains with small micro-magnetic resonance imaging reference datasets

The research focuses on automating the identification and segmentation of blood vessels in mice brains using micro-magnetic resonance imaging (μMRI). Manual segmentation is time-consuming, and automated methods often require substantial manual input. The study introduces a shallow, three-dimensional U-Net architecture for vessel segmentation that works with small datasets and requires only a small subset of labelled training data. This approach aims to improve the speed and reliability of vessel detection. The model's performance is evaluated using cross-validation, achieving an average Dice score of 61.34% in its best setup. The results indicate that the method detects blood vessels faster and more reliably compared to state-of-the-art vesselness filters.

• 1
  Deep learning architectures, specifically a shallow 3D U-Net model, are applicable for segmenting mice brain vessels using small μMRI reference datasets.
• 2
  The choice of post-processing methods, such as thresholding or region growing, has only a small influence on the final segmentation result.
• 3
  The deep learning-based methodology outperforms classic approaches like vesselness filters, achieving improved vessel detection rates.