Abstract:In spite of the great progress that has been made towards automating brain extraction in human magnetic resonance imaging (MRI), challenges remain in the automation of this task for mouse models of brain disorders. Researchers often resort to editing brain segmentation results manually when automated methods fail to produce accurate delineations. However, manual corrections can be labor-intensive and introduce interrater variability. This motivated our development of a new deep-learning-based method for brain segmentation of mouse MRI, which we call Mouse Brain Extractor. We adapted the existing SwinUNETR architecture ( ) with the goal of making it more robust to scale variance. Our approach is to supply the network model with supplementary spatial information in the form of absolute positional encoding. We use a new scheme for positional encoding, which we call Global Positional Encoding (GPE). GPE is based on a shared coordinate frame that is relative to the entire input image. This differs from the positional encoding used in SwinUNETR, which solely employs relative pairwise image patch positions. GPE also differs from the conventional absolute positional encoding approach, which encodes position relative to a subimage rather than the entire image. We trained and tested our method on a heterogeneous dataset of N=223 mouse MRI, for which we generated a corresponding set of manually-edited brain masks. These data were acquired previously in other studies using several different scanners and imaging protocols and included and images of mice with heterogeneous brain structure due to different genotypes, strains, diseases, ages, and sexes. We evaluated our method’s results against those of seven existing rodent brain extraction methods and two state-of-the art deep-learning approaches, nnU-Net ( ) and SwinUNETR. Overall, our proposed method achieved average Dice scores on the order of 0.98 and average HD95 measures on the order of 100 µm when compared to the manually-labeled brain masks. In statistical analyses, our method significantly outperformed the conventional approaches and performed as well as or significantly better than the nnU-Net and SwinUNETR methods. These results suggest that Global Positional Encoding provides additional contextual information that enables our Mouse Brain Extractor to perform competitively on datasets containing multiple resolutions.

Automated Skull Stripping in Mouse Functional Magnetic Resonance Imaging Analysis Using 3D U-Net

Automated 3D Fetal Brain Segmentation Using an Optimized Deep Learning Approach

Automatic Skull Stripping of Rat and Mouse Brain MRI Data Using U-Net

3D U-Net Improves Automatic Brain Extraction for Isotropic Rat Brain Magnetic Resonance Imaging Data

Fast and Lightweight Network Improves Serial Brain Section Stitching

Automated joint skull-stripping and segmentation with Multi-Task U-Net in large mouse brain MRI databases

A deep learning based tool for automatic brain extraction from functional magnetic resonance images in rodents

RS2-Net: An end-to-end deep learning framework for rodent skull stripping in multi-center brain MRI

Automatic Skull Stripping Using Multidimensional Multi-input Multi-output U-Net Model for Alzheimer's Disease

A Template‐based Automatic Skull‐stripping Approach for Mouse Brain MR Microscopy

Artificial Intelligence in Fetal Resting-State Functional MRI Brain Segmentation: A Comparative Analysis of 3D UNet, VNet, and HighRes-Net Models

Fully automated whole brain segmentation from rat MRI scans with a convolutional neural network

Automated registration-based skull stripping procedure for feline neuroimaging

Automatic fetal brain extraction from 2D in utero fetal MRI slices using deep neural network

Mouse Brain Extractor: Brain segmentation of mouse MRI using global positional encoding and SwinUNETR

Fully Automated Skull Stripping from Brain Magnetic Resonance Images Using Mask RCNN-Based Deep Learning Neural Networks

Dimensionality Reduction Hybrid U-Net for Brain Extraction in Magnetic Resonance Imaging

DIKA-Nets: Domain-invariant knowledge-guided attention networks for brain skull stripping of early developing macaques

Generalizing deep learning brain segmentation for skull removal and intracranial measurements.

Accurate brain extraction on MRI using U-Net trained in two stages

A prior feature SVM-MRF based method for mouse brain segmentation.