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

Abstract

Accurate removal of magnetic resonance imaging (MRI) signal outside the brain, a.k.a., skull stripping, is a key step in the brain image pre-processing pipelines. In rodents, this is mostly achieved by manually editing a brain mask, which is time-consuming and operator dependent. Automating this step is particularly challenging in rodents as compared to humans, because of differences in brain/scalp tissue geometry, image resolution with respect to brain-scalp distance, and tissue contrast around the skull. In this study, we proposed a deep-learning-based framework, U-Net, to automatically identify the rodent brain boundaries in MR images. The U-Net method is robust against inter-subject variability and eliminates operator dependence. To benchmark the efficiency of this method, we trained and validated our model using both in-house collected and publicly available datasets. In comparison to current state-of-the-art methods, our approach achieved superior averaged Dice similarity coefficient to ground truth T2-weighted rapid acquisition with relaxation enhancement and T2 *-weighted echo planar imaging data in both rats and mice (all p < 0.05), demonstrating robust performance of our approach across various MRI protocols.

Atlas-Based Brain Extraction Is Robust Across RAT MRI Studies

“…The CAMRI dataset (Accession Number ds002870, version 1.0.0, [8]) combines various studies of T2w RARE images from 94 Sprague Dawley, 22 Long-Evans, and 16 Wistar male rats. Although the acquisition parameters are not available, the image resolution for 69 rats was 0.1×0.1×1 mm with a 256×256×12 matrix, covering a limited FOV that excludes the olfactory bulb and cerebellum.…”

“…An alternative approach to atlas-based brain extraction, using convolutional neural networks, appears close to a breakthrough. Although results are promising on human adults (e.g., github.com/neuronets/nobrainer,) the limited availability of annotated rodent MRI data required to train the models precludes its generalization across studies [3,8].…”

“…This is problematic for rodent data, which differ in anatomy and various aspects of MRI acquisition. Brain extraction tools made specifically for rodent imaging have attempted to address these issues [3][4][5][6][7][8]. These methods mostly depend on the shape of the brain as defined by the image contrast, so their success is constrained to images acquired with standard parameters and with uniform signal.…”

3D U-Net improves automatic brain extraction for isotropic rat brain MRI data

“…To date, the most prominent tools to address rodent MRI brain extraction include Pulse-Coupled Neural Network (PCNN)-based brain extraction proposed by [8], Rapid Automatic Tissue Segmentation (RATS) proposed by [9], and SHape descriptor selected External Regions after Morphologically filtering (SHERM) proposed by [10], as well as a convolutional deep-learning based algorithm, 2D U-Net, proposed by [3]. While PCNN, RATS, and SHERM have demonstrated remarkable success (detailed introduction and comparisons discussed in [3]), their performance is subject to brain size, shape, texture, and contrast; hence, their settings often need to be adjusted per MRI-protocol for optimal results. In contrast, the U-Net algorithm explores and learns the hierarchical features from the training dataset, and provides a user-friendly and more universally applicable platform [11,12].…”

Automatic segmentation of the rat brain hippocampus in MRI after traumatic brain injury

“…Recently, CNNs have been successfully applied to a number of medical imaging segmentation tasks. For murine MRI, neural networks have been proposed for skull-stripping [10,11], lesion segmentation [12,13] and region segmentation [14].…”