Desikan-Killiany-Tourville Atlas Compatible Version of M-CRIB Neonatal Parcellated Whole Brain Atlas: The M-CRIB 2.0

Our recently published M-CRIB atlas comprises 100 neonatal brain regions including 68 compatible with the widely-used Desikan-Killiany adult cortical atlas. A successor to the Desikan-Killiany atlas is the Desikan-Killiany-Tourville atlas, in which some regions with unclear boundaries were removed, and many existing boundaries were revised to conform to clearer landmarks in sulcal fundi. Our first aim here was to modify cortical M-CRIB regions to comply with the Desikan-Killiany-Tourville protocol, in order to offer: (a) compatibility with this adult cortical atlas, (b) greater labeling accuracy due to clearer landmarks, and (c) optimisation of cortical regions for integration with surface-based infant parcellation pipelines. Secondly, we aimed to update subcortical regions in order to offer greater compatibility with subcortical segmentations produced in FreeSurfer. Data utilized were the T2-weighted MRI scans in our M-CRIB atlas, for 10 healthy neonates (post-menstrual age at MRI 40–43 weeks, four female), and corresponding parcellated images. Edits were performed on the parcellated images in volume space using ITK-SNAP. Cortical updates included deletion of frontal and temporal poles and ‘Banks STS,’ and modification of boundaries of many other regions. Changes to subcortical regions included the addition of ‘ventral diencephalon,’ and deletion of ‘subcortical matter’ labels. A detailed updated parcellation protocol was produced. The resulting whole-brain M-CRIB 2.0 atlas comprises 94 regions altogether. This atlas provides comparability with adult Desikan-Killiany-Tourville-labeled cortical data and FreeSurfer-labeed subcortical data, and is more readily adaptable for incorporation into surface-based neonatal parcellation pipelines. As such, it offers the ability to help facilitate a broad range of investigations into brain structure and function both at the neonatal time point and developmentally across the lifespan.


INTRODUCTION
We recently published the M-CRIB (Alexander et al., 2017) neonatal parcellated brain atlas, comprising 100 regions in total, including 68 compatible with the Desikan-Killiany (DK; Desikan et al., 2006) adult cortical atlas, as well as basal ganglia, thalamus, and cerebellar regions. The DK atlas is one of the most commonly used parcellation schemes, thus an advantage of the M-CRIB atlas is that it provides compatibility of parcellated cortical regions between neonatal and later time points. This can help facilitate investigations into regional brain structure and function across the lifespan, potentially longitudinally. As we have discussed previously (see Alexander et al., 2017), in neonates, the major sulci and gyri seen in adults are present at term (Chi et al., 1977;Cowan, 2002;Griffiths, 2010;Hill et al., 2010). The M-CRIB atlas illustrates the ability of the DK cortical regions to be delineated at the neonatal time point, as well as the subcortical and cerebellar regions; while also capturing gross and subtle morphological differences between neonatal and adult brains. The M-CRIB atlas is also valuable in that it comprises 10 individual high-quality detailed manual parcellations based on high resolution T2-weighted images, providing a combination of detailed whole-brain 'ground truth' and individual variability in morphology not available previously. We have recently demonstrated the applicability of the M-CRIB atlas, reporting differences in neonatal regional brain volumes based on premature birth (Alexander et al., 2018).
A successor to the DK atlas is the Desikan-Killiany-Tourville (DKT; Klein and Tourville, 2012) adult cortical parcellated atlas, in which some regions with unclear or arbitrary boundaries were removed, and many existing boundaries were revised to conform to sulcal fundi. This provides greater anatomical consistency across individuals due to clearer and more reproducible landmarks. The use of sulcal-based landmarks also optimizes utility for application using surface-based labeling such as is performed in FreeSurfer (Fischl et al., 2002). Surface-based methods incorporate surface-based registration which aligns sulci and gyri more precisely than volumebased methods (Fischl et al., 1999;Makropoulos et al., 2018), thus facilitating more precise alignment of sulcally-bounded labels.
A key resource facilitating accurate surface-based parcellation at the neonatal time point is high-quality ground truth neonatal parcellated training data. Such data are currently in strong demand.
Here we firstly aimed to modify the cortical regions and protocol of the existing volumetric M-CRIB atlas to comply with the DKT cortical parcellation protocol, in order to (a) offer compatibility with data at older time points parcellated with the adult DKT atlas, (b) achieve greater anatomical consistency in labeling across brains due to some boundaries being revised to clearer landmarks in sulcal fundi, and (c) offer greater ease of adaptability for integration into neonatal surface-based parcellation pipelines due to the use of these sulcallydefined boundaries. Secondly, we aimed to update subcortical regions to offer greater compatibility with those segmented by FreeSurfer's subcortical pipeline, including addition of the 'ventral diencephalon, ' and removal of 'subcortical matter' labels. These cortical and subcortical updates together comprise the 'M-CRIB 2.0' neonatal atlas.
T1-weighted images are included in the M-CRIB and M-CRIB 2.0 datasets, however, they were not used for manual tracing, because of low contrast between tissue types due to partial myelination at the neonatal time point. Rather, they are included as they may provide additional intensity information leverageable in multimodal automated parcellation pipelines. The T2-weighted images, which confer higher tissue contrast, were used both for parcellation of the original M-CRIB, and for the edits performed here.

Manual Editing Procedure
The individual segmentation images comprising the M-CRIB atlas were edited in volume space using Insight Toolkit (ITK)-SNAP v3.6.0 2 (Yushkevich et al., 2006), by one operator (B.A.). ITK-SNAP displays axial, sagittal, and coronal views and a composite 3D mesh representation of utilized labels. The edits were performed and checked on a combination of the axial, sagittal, and coronal views, with reference to the 3D surface view. Edits were performed region-by-region rather than brainby-brain to maximize consistency of labeling for each region across brains. An exception to this was in some areas where edits to multiple adjacent regions were required, as the alterations to one region sometimes necessitated specification of adjacent areas' boundaries. For some regions such as the newly-specified ventral diencephalon, edits were performed for the whole sample, and then checked and edited where necessary to ensure consistency.

Parcellation Protocols
In the following cortical protocol, revised boundary descriptors are listed that aimed to replicate the DKT (Klein and Tourville, 2012) protocol as closely as possible within this volumetric neonatal sample. Where possible, verbatim DKT boundary descriptors have been utilized, and are indicated in bold font. Descriptors retained from the DK protocol are indicated in italics. Descriptors either retained from the M-CRIB protocol or newly specified here are indicated in regular font. Some anatomical axis descriptors (e.g., 'anterior') have been adjusted to retain anatomical accuracy in volume space.
Updates between the M-CRIB and M-CRIB 2.0 atlases pertain to the DKT cortical regions, ventral diencephalon (added), brainstem (edited in the course of defining ventral diencephalon), left and right 'subcortical matter' (removed), and left and right cerebral white matter (edited in removal of subcortical matter labels). Cerebellum, hippocampus, amygdala, and ventricles, were retained as per the original M-CRIB atlas, and parcellation protocols for these regions are listed in Loh et al. (2016) and Alexander et al. (2017). Basal ganglia and thalamus were not manually edited and protocols for these regions are retained from the M-CRIB atlas, however, post-processing performed on these segmentations was removed, as described below.

Cortical Regions
Frontal pole, temporal pole, and "banks of the superior temporal sulcus" regions were removed as per the DKT protocol, and replaced with surrounding gyral labels.

www.braininfo.org
Medial: Medial aspect of the temporal lobe. Lateral: Rhinal sulcus (collateral sulcus), or the collateral sulcus if the rhinal sulcus is not present.
Parahippocampal gyrus Boundaries: Anterior: Posterior limit of the amygdala. Posterior: Posterior limit of the hippocampus. Medial: Medial aspect of the temporal lobe. Lateral: Collateral sulcus.
Temporal pole (removed): The area included in the DK temporal pole has been redistributed to the superior, middle and inferior temporal gyrus regions.
Fusiform Gyrus Boundaries: Anterior: Anterior limit of occipitotemporal sulcus (anterior limit of collateral sulcus). Posterior: First transverse sulcus posterior to the temporooccipital notch. This is consistent with the posterior extent of the existing parcellation, which was based on the M-CRIB boundary listed as "posterior transverse collateral sulcus (Duvernoy et al., 1999)." Medial: Collateral sulcus. Lateral: Occipitotemporal sulcus.

Temporal -lateral aspect
Superior temporal gyrus Boundaries: Anterior: Anterior limit of the superior temporal sulcus or a projection from the superior temporal sulcus to the anterior limit of the temporal lobe. Posterior: Junction of posterior horizontal ramus of the lateral sulcus (or its posterior projection) and caudal superior temporal sulcus (1 st segment of the caudal superior temporal sulcus). Note: The DKT protocol lists 1 st , 2 nd , or 3 rd segment, however, the current parcellations of this region posteriorly conform specifically to the landmark that Petrides (2011) describes as the 1 st segment, i.e., bounding the posterior extent of supramarginal gyrus (Petrides, 2011). Superomedial: Lateral fissure (and when present, the supramarginal gyrus and insula) Inferior: Superior temporal sulcus.
Middle temporal gyrus Boundaries: Anterior: Anterior limit of the superior temporal sulcus. Posterior: Anterior occipital sulcus. Note: this has also been described as the ascending limb of inferior temporal sulcus (Watson et al., 1993;Dumoulin et al., 2000;Petrides, 2011). This is described by Duvernoy et al. (1999) as only sometimes being present: "The inferior temporal sulcus is usually not continuous and does not provide easy identification. In the vicinity of the occipital lobe, its posterior end may occasionally run upward and be called the anterior occipital sulcus." In cases where this sulcus segment did not occur, the boundary was a point on a theoretical line extending vertically from the occipito-temporal incisure on the cortical surface. Superomedial: Superior temporal sulcus anteriorly, posteriorly formed by caudal superior temporal sulcus third segment. Inferior: Inferior temporal sulcus.
Inferior temporal gyrus Boundaries: Anterior: Anterior limit of the inferior temporal sulcus. Posterior: Anterior occipital sulcus (see descriptor for posterior boundary of middle temporal gyrus). In cases where this sulcus segment did not occur, the boundary was a point on a theoretical line extending vertically from the occipito-temporal incisure on the cortical surface. Superior: Inferior temporal sulcus Inferior: Occipitotemporal sulcus (Duvernoy et al., 1999).
Transverse temporal cortex Description: Also termed Heschl's gyrus, this area lies along the superior temporal plane, extending from the retroinsular region to the lateral edge of the superior temporal gyrus. It can be a single gyrus, or divided into two gyri by an intermediate transverse temporal sulcus (Duvernoy et al., 1999;Rademacher, 2003). Boundaries: Anterior: Anterior limit of first transverse temporal sulcus (also referred to as the anterior transverse temporal sulcus (Tamraz and Comair, 2006).) Posterior: Posterior limit of Heschl's sulcus [also referred to as the posterior transverse temporal sulcus (Rademacher, 2003;Tamraz and Comair, 2006), or transverse temporal sulcus (Duvernoy et al., 1999;Ono et al., 1990)]. Medial: Retro-insular area of the lateral fossa. Lateral: Lateral surface of the superior temporal gyrus. Inferior frontal gyrus Description: The inferior frontal gyrus comprises the three pars regions.
Inferior frontal gyrus -pars triangularis Boundaries: Anterior: Pretriangular sulcus. Posterior: Anterior ascending ramus of the lateral sulcus. Superomedial: Inferior frontal sulcus. Inferomedial: Anterior horizontal ramus of the lateral sulcus; if the anterior horizontal ramus of the lateral sulcus does not extend anteriorly to pretriangular sulcus, an anterior projection from anterior horizontal ramus of the lateral sulcus to pretriangular sulcus.
Inferior frontal gyrus -pars orbitalis Boundaries: Anterior: Pretriangular sulcus -if pretriangular sulcus does not extend ventrally to the lateral H-shaped orbital sulcus, a ventral projection from pretriangular sulcus to lateral H-shaped orbital sulcus completes the anterior boundary.  Ventral: Dorsal to the corpus callosum, the ventral boundary is formed by the callosal sulcus. In the subgenual area, it is formed by the cingulate sulcus. In the case of "double parallel cingulate" sulcus that continues anteroventrally to join the 'superior rostral sulcus' (listed in Klein and Tourville, 2012), the ventral boundary is the superior rostral sulcus, also termed 'supraorbital sulcus' (Duvernoy et al., 1999, p. 33 (Duvernoy et al., 1999) if present, or the parieto-occipital fissure. Lateral: The depth of the calcarine sulcus.

Subcortical Regions Basal ganglia and thalamus
The manual tracing protocol for the M-CRIB basal ganglia nuclei (caudate, putamen, pallidum, and nucleus accumbens) and thalamus is described in Loh et al. (2016). These regions   were not manually edited here. However, in the original M-CRIB dataset, basal ganglia and thalamus segmentations underwent morphological smoothing. For the M-CRIB 2.0, the smoothed segmentations of these structures were replaced with nonsmoothed segmentations in order to recover fine-scale, irregular, intensity-based anatomical detail such as is provided for the rest of the M-CRIB and M-CRIB 2.0 regions.

Ventral diencephalon
The protocol for this region is based on that of de Macedo Rodrigues et al. (2015). Boundaries: Anterior: Anterior commissure (however, unlike the protocol of de Macedo Rodrigues et al., 2015, where the infero-rostral boundary is designated as the infundibular recess, we have referred solely to the anterior commissure as an anterior boundary, as much of the optic recess was also visible posterior to the anterior commissure). Posterior: Medially, the posterior commissure. Laterally, the posterior extent of the lateral geniculate nucleus. However, the lateral geniculate nucleus itself was retained as part of the thalamus label. Superior: The inferior surface of the thalamus, posteriorly (as per de Macedo Rodrigues et al., 2015).
Inferior: A line extending from the pontomesencephalic sulcus anteriorly, to the posterior commissure posteriorly. Lateral: The optic pathways (de Macedo Rodrigues et al., 2015).

Brainstem
The M-CRIB brainstem label was originally derived via the initial automated MANTiS (Beare et al., 2016) tissue segmentation, and refined during the process of manually delineating surrounding structures. Here partial sections of the cerebral peduncles, red nucleus, and substantia nigra have been reassigned from the brainstem label to form part of the ventral diencephalon label.

RESULTS
The M-CRIB 2.0 atlas comprises 94 regions: 62 cortical regions, and subcortical and cerebellar regions from the M-CRIB atlas. Figures 1, 2 illustrate some of the updates made, displayed on surface meshes and axial slices, respectively. Atlas colors and corresponding label names are shown in Supplementary Figure S1.
In Figure 2, altered regions surrounding basal ganglia and thalamus primarily reflect the removal of the 'subcortical matter' label, which was replaced with 'ventral diencephalon' and cerebral white matter labels. Table 1 lists the mean volume of each M-CRIB 2.0 region, and the volume relative to the equivalent structure, where applicable, from the original M-CRIB atlas.

DISCUSSION
The M-CRIB and M-CRIB 2.0 atlases provide neonatal parcellated regions compatible with those in adults, while also representing the gross and subtle morphological differences between neonatal and adult brains. The adult DK and DKT cortical atlases are comprised of major gyri and large-scale regions that are extant at term in infants. The DKT protocol involved removal of some abstractly-bounded regions, and specification of additional sulci as regional boundaries. These sulci were generally readily identifiable in the current data. In a few instances, however, boundaries consisting of minor sulci were not identifiable in the neonatal data. In these cases boundaries were adjusted to the most closely-equivalent boundary in the neonatal data to correspond with that specified in adults. For example, the posterior boundary of the supramarginal gyrus was defined as the first segment of the caudal superior temporal sulcus in the neonatal data, because the second and third segments (specified as alternatives in the adult DKT protocol) were not consistently identifiable. Such differences reflect morphology specific to neonates, highlighting the value of delineating atlases in neonatal data, rather than projection of atlases defined in adults into neonatal space. Similarly, a small number of non-cortical regions in the M-CRIB and M-CRIB 2.0 atlases are necessarily inconsistent between neonates and adults. For example, cerebellar white and gray matter can be parcellated separately in adults. However, these finelyinterbranched structures are not delineable in neonatal data at the current resolution due to partial voluming, so cerebellar hemispheres are provided as single structures. Conversely, the CSF label in neonates includes the cavum septum pellucidumthe CSF-filled space between the two septa pellucida, however, this structure is infrequently seen in adults (Tubbs et al., 2011). Such differences necessitate clear protocol descriptions in neonatal data as we have provided, to facilitate clear understanding of the parcellated anatomy. The M-CRIB 2.0 atlas incorporates updates that increase compatibility with adult subcortical segmentations derived via FreeSurfer, namely the addition of ventral diencephalon, and removal of 'subcortical matter' labels.
The parcellated M-CRIB 2.0 images will be more readily adaptable for potential incorporation into surface-based neonatal parcellation pipelines. Indeed, forthcoming work from our lab consists of the production of surface-based templates of the DKcompatible M-CRIB cortical regions, and the DKT-compatible M-CRIB 2.0 cortical regions. Accompanying the templates will be a protocol for labeling neonatal data using these atlases in combination with existing infant surface-based tools.
The current volumetric parcellations may also be used in combination with labeling tools such as STAPLE (Zou et al., 2004;Akhondi-Asl and Warfield, 2013;Akhondi-Asl et al., 2014) that apply labels to new data in a probabilistic, voxelwise manner (see Alexander et al., 2018, for an example).
The individual volumetric parcellated images and T1-and T2weighted images comprising the M-CRIB 2.0 atlas will be publicly available.

CONCLUSION
We updated the M-CRIB neonatal parcellated brain atlas to be compatible with the DKT adult cortical parcellated atlas, and to incorporate updates to subcortical regions facilitating greater compatibility with FreeSurfer's subcortical segmentation.
We achieved this via manual volumetric edits to the individual parcellated images, and via the production of a detailed, revised whole-brain parcellation protocol. The resulting M-CRIB 2.0 atlas offers greater compatibility with adult parcellated data, greater accuracy due to more reproducible landmarks, and greater optimisation for integration with surface-based infant cortical parcellation pipelines. This high-quality dataset can therefore help facilitate a broad range of investigations into brain structure and function both at the neonatal time point and developmentally across the lifespan.

ETHICS STATEMENT
This study was carried out in accordance with the recommendations of the National Statement on Ethical Conduct in Human Research (2007), and Royal Children's Hospital Human Research Ethics Committee, with written informed parental consent provided for all subjects. Written informed parental consent was given in accordance with the Declaration of Helsinki. The protocol was approved by the Royal Children's Hospital Human Research Ethics Committee.

AUTHOR CONTRIBUTIONS
BA, WL, LM, and AM performed manual parcellation of the original atlas. CA, RB, and JC provided technical and methodological support in its development. PA, LD, AS, JLYC, MS, and DT facilitated collection of the data. DT contributed conception of the atlas. BA performed edits to the data and wrote the manuscript. CK provided conceptual consultation. All authors contributed to manuscript revision, read and approved the submitted version.