Table 1 Review of available neonatal and infant templates and atlases
From: The FinnBrain multimodal neonatal template and atlas collection
Study | N | M | L | K | R | A | S |
|---|---|---|---|---|---|---|---|
Akiyama Atlas (Akiyama et al., 2013) | 60 (29 m, 31 f) 1.5 T scanner (N = 27; 14 m, 13 f), 3 T scanner (N = 33; 15 m, 18 f) | sMRI (1.5 T scanner: T1, images only in sagittal plane), sMRI (3T scanner:T1) | 90 (AAL) | 1 | 1.5 T scanner 0.94 × 0.94 × 1 mm3; 3 T scanner 1 × 1 × 1 mm3 | mean = 204.0 ± 12.2 days (range = 177–230), median = 202.5 days, | Extracted 6-month-old average brain was segmented into brain tissue and CSF using FSL FMRIB’s automated segmentation tool (FAST); ANTS. Template construction using tools from the Medical Image NetCDF (http://www.bic.mni.mcgill.ca/ServicesSoftware). |
Altaye Atlas (Altaye et al., 2008) | 77 (31 m, 46 f) | sMRI (T1) | 3 (GM, WM, CSF) | 1 | 1 × 1 × 1 mm3 | range = 9–15 months | Dual strategy segmentation approach using SPM5: i) unified segmentation based on a priori adult segmentations ii) current voxel-intensity approach based on a Gaussian mixture model. The final tissue probabilities are estimated without tissue priors. Both strategies apply an HMRF model (Cuadra et al., 2005) during template construction. |
EBDS Neonatal DTI Atlas (Short et al., 2022) | Newborn: 144 (68 m, 76 f); 1-year-old: 170 (95 m,75 f); 2-year-old: 171 (88 m,83 f) | DTI | 47. | 3 | 2 × 2 × 2 mm3 | Newborn: mean = 41.88 ± 1.83 (range = 38.14–48) gestational weeks; 1-year-old: mean = 56.6 ± 0.59 (range = 47.43–73) weeks; 2-year-old: mean = 145.82 ± 3. (range = 94.43–121.14) weeks | Fibre-tract-based analysis closely follows UNC Utah NAMIC DTI analysis framework (Verde et al., 2013, 2014) https://www.nitrc.org/projects/namicdtifiber/DTIAtlasBuilder). Atlas construction using DTIAtlasBuilder (https://www.nitrc.org/projects/dtiatlasbuilder) |
Edinburgh Neonatal Atlas (ENA33) (Blesa et al., 2016) https://git.ecdf.ed.ac.uk/jbrl/ena/tree/00fe2c0ae25f326338369175643356acf272f780 | 33 | sMRI (T1, T2), DTI | 107 | 1 | T1: 1 × 1 × 1 mm3; T2: 0.9 × 0.9 × 0.9 mm3; DTI: 2 × 2 × 2 mm3 | mean post-menstrual age = 42 (range = 39–47) weeks | Tissue segmentation used non-linear registration to the closest age-matched T1w template from the 4D atlas using Free-Form Deformation implemented in NiftyReg, expectation-maximisation (EM) algorithm used to classify each voxel into tissue. The neonatal brain was parcellated into the SRI24/TZO adult brain atlas (Rohlfing et al., 2009) using the LISA method (Serag et al., 60) to model the anatomical differences between adult and neonatal brains. Template and atlas were constructed using the Symmetric Group Normalisation (SyGN) method. |
Imperial ALBERTs (Gousias et al., 2012) https://brain-development.org/brain-atlases/neonatal-brain-atlases/neonatal-brain-atlas-gousias/ | 20: 15 preterm (7 m, 8 f); 5 term-born (3 m, 2 f) | sMRI (T1,T2) | 50 | 1 | T1: 0.82 × 0.82 × 1.6 mm3 (resliced to 0.82 × 0.82 × 0.82 mm3); T2: 0.86 × 0.86 × 2.0 mm3 | preterm: median post-menstrual age = 40 (range = 37–43) weeks; term-born: median post-menstrual age = 41 (range = 39–45) weeks | Manual segmentation of the whole brain into 50 regions based on previous protocols (Ahsan et al., 2007; Gousias et al., 2008; Hammers et al., 2003, 2007) using macroanatomical landmarks. Each voxel was labelled as belonging to one ROI, resulting in a label-based encephalic ROI template (ALBERT). |
Imperial Neonatal Atlas (Kuklisova-Murgasova et al., 2011) https://brain-development.org/brain-atlases/neonatal-brain-atlases/neonatal-brain-atlas-murgasova/ | 142 (70 m, 72 f) | sMRI (T2) | 6 | 1 | 0.86 × 0.86 × 1 mm3 | mean = 36.6 ± 4.9 gestational weeks (range 28.6–47.7) | Brain segmentation with an intensity-based approach similar to the work of (Xue et al., 2007), using atlas-based segmentations based on manual delineations of deep grey matter, brainstem, cerebellum and darker regions of white matter. Kernel-based regression method was used for template and age-specific 4D probabilistic atlas creation (Davis et al., 2010; Ericsson et al., 2008). |
Imperial Paediatric Atlas (Gousias et al., 2008) https://brain-development.org/brain-atlases/pediatric-brain-atlases/pediatric-brain-atlas-gousias/ | 33 (17 m, 16 f) | sMRI (T1, images only in sagittal plane) | 83 | 1 | 1.04 × 1.04 × 1.04 mm3 | mean = 24.8 ± 2.4 (range 21.4–34.4) months, median = 24.1 months | Automatic segmentation of paediatric brains using an algorithm that was based on manual segmentation of 30 adult brains that resulted in 30 adult atlases labelling 83 anatomical structures. Final segmentation combined the 30 adult atlases using decision fusion. |
Imperial SpatioTemporal Atlas (Serag et al.60) https://brain-development.org/brain-atlases/neonatal-brain-atlases/neonatal-brain-atlas-serag/ | 204 | sMRI (T1,T2) | 6 | 1 | T1: 0.82 × 1.03 × 1.6 mm3; T2: 1.15 × 1.18 × 2 mm3 | mean = 37.3 ± 4.8 post-menstrual weeks (range = 26.7–44.3) | Procedure following Imperial Neonatal Atlas (Kuklisova-Murgasova et al., 2011). |
Infant FreeSurfer Atlases (de Macedo Rodrigues, 2015) | 23 (8 m, 15 f) | sMRI (T1) | 32 + 14 | 1 | 1 × 1 × 1 | range = 0 –18 months | Manual segmentation |
INSERM Atlas (Dehaene-Lambertz et al., 2002) | 20 (6 m, 24 f) | sMRI (T2) | 13 | 1 | 0.391 × 0.391 × 4 mm3 (resampled at 3.1 × 3.1 × 4 mm3) | range = 2 –3 months | The template was constructed using manual alignment of the AC-PC commissures for two participants using SPM99 and Anatomist. |
JHU-neonate-linear and JHU-neonate-non-linear-atlases (Oishi et al., 2011) https://cmrm.med.jhmi.edu/cmrm/Data_neonate_atlas/atlas_neonate.htm | T1 (N = 14), T2 and DTI (N = 20) | sMRI (T1, T2), DTI | 122 | 1 | T1, T2: 1 × 1 × 1 mm3; DTI: 0.6 × 0.6 × 0.6 mm3 | Term-born, 0–4 days after birth | Manual segmentation procedure follows the adult JHU-MNI parcellation map. Each T2w image was aligned to a common AC-PC line based on the averaged brain size and shape. This was applied to all collected modalities creating the JHU-neonate-linear atlas. Coregistered DTI images were nonlinearly normalised to the JHU-neonate-SS atlas with a dual-channel LDDMM (Ceritoglu et al., 2009; Miller et al., 1997) using DiffeoMap (see (Oishi et al., 2009). The resulting non-linear transformation matrices were applied to the coregistered T1w and T2w images creating the JHU-neonate-non-linear-atlases. |
JHU-neonate-SS Atlas (Oishi et al., 2011) https://cmrm.med.jhmi.edu/cmrm/Data_neonate_atlas/atlas_neonate.htm | 1 | sMRI (T1, T2) | 122 | 1 | 1 × 1 × 1 mm3 | Term-born, 2 days after birth | Manual segmentation procedure follows the adult JHU-MNI parcellation map using ROIEditor (www.Mristudio.org) to inspect all three slice orientations. One subject with the best fitting brain shape to the JHU-neonate-linear-atlas was linearly normalised to the T2w image of the JHU-neonate-linear. The resulting transformation matrix was applied to the other coregistered DTI and T1w images creating the JHU-neonate-SS. |
M-CRIB Atlas (Alexander et al., 2017) https://github.com/DevelopmentalImagingMCRI/M-CRIB_atlas M-CRIB 2.0 | 10 (6 m, 4 f) | sMRI (T2) | 100 | 1 | 0.63 × 0.63 × 0.63 mm3 | Term-born, mean = 41.71 (range 40.29–43) gestational weeks | MANTiS for automatic tissue classification, then manual cleaning of tissue segmentation; all parcellation of high-resolution T2w images using ITK-SNAP (Yushkevich et al., 2006) and manual parcellation for 100 different regions. Linear and non-linear T1w and T2w templates were constructed using ANTS V2.1. |
MRICloud neonate multi-atlas repository (Otsuka et al., 2019) | 7 (3 m, 4 f) | sMRI (T1) | 30 | 7 | 1 × 1 × 1 mm3 | Term- and preterm-born, range = 38–42 weeks | Automatic parcellation using MALF integrated with segmentation tools in MRICloud (https://mricloud.org/). |
Multi-structural Neonatal Brain Atlas (Makropoulos et al. 2016) | 338 (298 preterm) | sMRI (T2) | 50 (82) | 5 | 0.86 × 0.86 × 2 mm3 | Term-born: mean = 0 (range = 0–5) weeks; preterm: mean = 6 (range 0–19) weeks | Segmentation protocol following (Makropoulos et al., 2014); Expectation-maximization algorithm; image intensity modelled with Gaussian Mixture Model. A 4D spatiotemporal structural atlas of the brain built from the 82 cortical and subcortical segmentation averages. |
NIHPD Objective 2 Atlases (Fonov et al. 2009) | 108 | sMRI (T1) | No anatomical parcellation provided | 11 | 1 × 1 × 3 mm3 | range = 0–4.5 years | A suite of software developed by the Montreal Neurological Institute (MNI) was used. PMID N/A https://www-sciencedirect-com.ezproxy.utu.fi/science/article/pii/S1053811909708845 |
Singapore Atlas (Broekman et al. 2014) | 93 (44 m, 49 f) | sMRI (T2), DTI | 2 | 2 | 3.125 × 3.125 × 3 mm3 | mean = 9.9 ± 2.3 (range = 5–17) days | DTI Atlas was created using the unbiased diffeomorphic atlas generation algorithm (Joshi et al., 2004). FA image aligned to JHU-neonate-SS DTI atlas (Oishi et al., 2011); Voxel-based analysis using SPM8. |
UNC-CH Longitudinal Infant Atlas (Kim et al. 2017) | 8 | DWI | 1 | 3 | 2 × 2 × 2 mm3 | Term-born, neonate, 6 months and 12 months | All DW images were processed using FSL. DW atlases were constructed by fusing diffusion-weighted images across time points and space in a patch-wise way using sparse representation with a graph constraint that promotes spatiotemporal consistency. |
UNC-CH Neonatal Atlas (Saghafi et al. 2017) | 30 | DWI | 2 | 1 | 2 × 2 × 2 mm3 | 14 days | Atlas was constructed with a patch-based method, that jointly considers neighbouring gradient directions in the DW images. A group regularisation framework is used to constrain local patches for consistent spatio-angular reconstruction. |
UNC Cortical (Li et al., 2015) | 35 participants (18 m, 17 f); 202 scans (4–7 per infant; the number of scans was N = 35 at 1 month, N = 28 at 3 months, N = 31 at 6 months, N = 27 at 9 months, N = 29 at 12 months, N = 31 at 18 months, and N = 21 at 24 months | sMRI (T1, T2), DWI | No anatomical parcellation provided | 7 | T1: 1 × 1 × 1 mm3 T2: 1.25 × 1.25 × 1.95 mm3 (resampled to 1 × 1 × 1 mm3) DWI: 2 × 2 × 2 mm3 (resampled to 1 × 1 × 1 mm3) | 1, 3, 6, 9, 12, 18, and 24 months | Volumetric segmentation in line with their prior work (Li et al., 2013), i.e. a longitudinally consistent tissue segmentation by an infant-dedicated, 4D level-set method referencing iBEAT software (Wang et al., 2011, 2012, 2014). Groupwise surface registration was used for the creation of the cortical surface atlas with Spherical Demons (Yeo et al., 2010). |
UNC detail-preserved longitudinal 0-3-6-9-12 months-old atlas (Zhang et al., 2016) | 35 (18 m, 17 f); 150 scans (2–5 per infant | sMRI (T1, T2, only images in sagittal plane) | 3 | 4 | T1: 1 × 1 × 1 mm3 T2: 1.25 × 1.25 × 1.95 mm3 (resampled to 1 × 1 × 1 mm3) | Term-born, range = 0–13 months years | Tissue segmentation was carried out with iBEAT. The template construction was carried out with a novel framework for consistent spatial-temporal construction of longitudinal atlases where the atlas construction was performed in spatial-temporal wavelet domain simultaneously. |
UNC/UCI neonate hippocampus amygdala multi-atlas | 6 | sMRI (T1, T2) | 7 | 1 | 1 × 1 × 1 mm3 | Term-born, 0–5 weeks | Manual segmentation protocol. Publication N/A |
UNC/UMN Baby Connectome Project (BCP) Atlases (Ahmad et al., 2023) | 37 (17 m, 20 f); 108 scans | sMRI (T1, T2) | 3 | 7 | 0.8 × 0.8 × 0.8 mm3 | 2 weeks, 3, 6, 9, 12, 18, and 24 months | Tissue segmentation using iBEAT V2.0 (https://ibeat.wildapricot.org). The 12-month surface-volume atlas was constructed using a dynamic elasticity model with surface constraint (SC-DEM) for groupwise registration of tissue segmentation maps. The 2 weeks–24 months longitudinal atlases were constructed using parallel transported longitudinal deformations. |
UNC volumetric/UNC-infant-0-1-2 atlases (Shi et al., 2011) http://bric.unc.edu/ideagroup/free-softwares/unc-infant-0-1-2-atlases/ | 95 (56 m, 39 f) | sMRI (T2 for neonates and T1 for 1- and 2-year-olds) | 90 | 3 | T1: 1 × 1 × 1 mm3 T2: 1.25 × 1.25 × 1.95 mm3 | Neonate: 41.5 ± 1.7 (range = 38.7–46.4) weeks; 1-year-old: 94.2 ± 3.4 (range = 87.9–109.1); 2-year-old: 146.2 ± 4.9 (range = 131.4–163.4) | ITK-SNAP (Yushkevich et al., 2006) was used for ground-truth manual segmentation of the neonates. SPM5 is used for atlas-based segmentation. A groupwise registration algorithm (Wu et al., 2011) was used for the atlas construction of each three age groups. |
USC atlas/template (Sanchez et al., 2012) Neurodevelopmental MRI Database (Richards et al. 2016) | Scan images obtained from two sources: NIHPD & MCBI; NIHPD = 105 (59 m, 46 f); MCBI = 49 (24 m, 25 f) | NIHPD- sMRI (T1, T2, only images in axial plane); MCBI- sMRI (T1 images in sagittal plane and T2 images in axial plane) | 3 | 13 | NIHPD- 1 × 1 × 1 mm3 ; MCBI- T1: 1 ×1 × 1 mm3 and T2: 1 × 1 × 1 to 2.5 mm3 | Range = 8 days–4.3 years (13 groups; mean ages 2 weeks, 3, 4.5, 6, 7.5, 9, 12, 15, 18 months, 2, 2.5, 3, 4 years) | FSL FLIRT was used to make a preliminary template of four 6-month-olds’ heads and brains from the USC-MCBI dataset; SPM8; and ANTS were used for template construction. |
Zhang DTI Atlas (Zhang et al. 2014) | 9 (2 m, 7 f) | sMRI (T1), DTI | 122 | 1 | T1: T1: 1 × 1 × 1 mm3; DTI: 2 × 2 × 2.5 mm3 | 2–13 days | The template was constructed using a volume-based template estimation (VTE) method. VTE was morphed to the JHU-neonate-SS atlas parcellation to label the anatomical features. |
