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

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

“…image noise, bias field in [21]) and no a-priori information are required (e.g. center of gravity distance map, signed distance map in [18]).…”

“…We complete our comparison with state of the art by positioning our performances in relation to methods applied to other animal models: macaques [18,13], rodents [17,23,24,13] and pigs [21]. This report of performance of related works is for information purposes since dataset are different and all third party codes are not freely available [18,21,23].…”

“…Thus, Ebner et al [15] proposes a pipeline for the reconstruction of the fetal brain starting with a skull stripping using 2D Convolutional Neural Network (CNN) while Salehi et al [16] considers brain segmentation with a 2D Unet. However, the use of 3D Convolutional neural networks for the segmentation of animal brains is still little exploited despite good results in rodents (PCNN3D [17]) or macaques (DIKA-Nets [18]). It has never been used to segment a piglet brain.…”

“…Compared to human brain, being a large structure surrounded by highly contrasted skull, one considers piglet brains, being a small structure surrounded by muscles and fats [12]. Concerning the brain segmentation of other animals, one can mention works related to pigs [21], macaques (DIKA-Net [18], atlasBREX [13]) and rodents (2D U-Net [23], PCNN3D [17], RATS [24] and atlasBREX). Our proposal differs from existing techniques because it considers a 3D patch-based approach (only 2D in [23], 3D without patches in [21,17]), it is atlas-free (unlike [13]) and addresses the issue of longitudinal monitoring (out of the scope of many related works [17,24,21,23]) by using transfer learning (ignored in [18]).…”

A Macaque Brain Extraction Model Based on U-Net Combined with Residual Structure

“…Brain tissue extraction focuses on removing non-brain tissues, such as the skull, muscles, and eyes, and then preserving the brain tissue [ 12 , 13 , 14 ]. Numerous software tools, for use on the human brain for skull stripping, have been developed, such as the Brain Extraction Tool (BET) in FSL [ 15 , 16 ], 3dSkullStrip in AFNI [ 17 , 18 , 19 ], and the hybrid watershed algorithm (HWA) in FreeSurfer [ 20 , 21 ]. Although these tools can perform well, when applied to the human brain, their performance is lacking when used on the macaque brain, mainly due to the image differences between the macaque and human brains [ 22 ].…”