Surface-based hippocampal subfield segmentation

Abstract

The human hippocampus is composed of a folded archicortical sheet with its subfields containing unique cellular compositions.Magnetic resonance imaging (MRI) is a promising way to study subfield functions and abnormalities in disease, but alignment to histologically defined subfield atlases is challenging due to extensive interindividual variability.Surface-based methods allow alignment of topologically homologous tissues between individuals or from one individual to a histological reference.Compared with manual or registration-based approaches, surface-based approaches provide new biologically valid constraints to subfield segmentation and do not suffer some of the technical limitations, such as out-of-plane sampling, of manual approaches.In particular, such methods show promise in high-resolution imaging where assessing interindividual variability within the hippocampus is becoming increasingly feasible. Though it is often termed ‘subcortical,’ the hippocampus is composed of a folded ‘archicortical’ sheet contiguous with the neocortex. The human hippocampus varies considerably in its internal folding configuration, creating major challenges in interindividual alignment and parcellation into subfields. In this opinion article, we discuss surface-based methods that aim to explicitly model hippocampal folding, similar to methods used in the neocortex, allowing interindividual alignment in an unfolded or flat-mapped 2D space. Such an approach enables detailed morphological characterization, constrains the problem of subfield segmentation, and provides a way to visualize data without occlusions. We argue that, when applied to magnetic resonance imaging (MRI) data, such methods overcome pitfalls of more conventional manual or registration-based subfield segmentation approaches. Though it is often termed ‘subcortical,’ the hippocampus is composed of a folded ‘archicortical’ sheet contiguous with the neocortex. The human hippocampus varies considerably in its internal folding configuration, creating major challenges in interindividual alignment and parcellation into subfields. In this opinion article, we discuss surface-based methods that aim to explicitly model hippocampal folding, similar to methods used in the neocortex, allowing interindividual alignment in an unfolded or flat-mapped 2D space. Such an approach enables detailed morphological characterization, constrains the problem of subfield segmentation, and provides a way to visualize data without occlusions. We argue that, when applied to magnetic resonance imaging (MRI) data, such methods overcome pitfalls of more conventional manual or registration-based subfield segmentation approaches. Hippocampal subfields are regions with stereotyped cellular composition that have come to the forefront of current research because of their promise in furthering both clinical and theoretical neuroscience research. Precise estimates of subfield locations can be used to test hypotheses about the relationship of subfield architecture and function via functional magnetic resonance imaging (fMRI) or other recording methods in humans [1.Duvernoy H.M. et al.The Human Hippocampus: Functional Anatomy, Vascularization and Serial Sections with MRI. Springer Science & Business Media, 2013Crossref Scopus (111) Google Scholar, 2.Olsen R.K. Robin J. Zooming in and zooming out: the importance of precise anatomical characterization and broader network understanding of MRI data in human memory experiments.Curr. Opin. Behav. Sci. 2020; 32: 57-64Crossref Scopus (6) Google Scholar, 3.Giuliano A. et al.Hippocampal subfields at ultra high field MRI: an overview of segmentation and measurement methods.Hippocampus. 2017; 27: 481-494Crossref PubMed Scopus (35) Google Scholar]. Differences in subfield-specific integrity have been observed in many diseases or disease subtypes (e.g., [4.Blümcke I. et al.Defining clinico-neuropathological subtypes of mesial temporal lobe epilepsy with hippocampal sclerosis.Brain Pathol. 2012; 22: 402-411Crossref PubMed Scopus (122) Google Scholar, 5.Steve T.A. et al.Hippocampal subfield measurement and ILAE hippocampal sclerosis subtype classification with in vivo 4.7 Tesla MRI.Epilepsy Res. 2020; 161: 106279Crossref PubMed Scopus (11) Google Scholar, 6.La Joie R. et al.Hippocampal subfield volumetry in mild cognitive impairment, Alzheimer’s disease and semantic dementia.Neuroimage Clin. 2013; 3: 155-162Crossref PubMed Scopus (158) Google Scholar, 7.Haukvik U.K. et al.Neuroimaging hippocampal subfields in schizophrenia and bipolar disorder: a systematic review and meta-analysis.J. Psychiatr. Res. 2018; 104: 217-226Crossref PubMed Scopus (61) Google Scholar]), and subfield-specific functions have been linked to distinct aspects of cognition (e.g., [8.Hunsaker M.R. Kesner R.P. The operation of pattern separation and pattern completion processes associated with different attributes or domains of memory.Neurosci. Biobehav. Rev. 2013; 37: 36-58Crossref PubMed Scopus (162) Google Scholar, 9.Rolls E.T. Kesner R.P. Pattern separation and pattern completion in the hippocampal system: introduction to the special issue.Neurobiol. Learn. Mem. 2016; 129: 1-3Crossref PubMed Scopus (9) Google Scholar, 10.Dimsdale-Zucker H.R. et al.CA1 and CA3 differentially support spontaneous retrieval of episodic contexts within human hippocampal subfields.Nat. Commun. 2018; 9: 294Crossref PubMed Scopus (78) Google Scholar]). Thus, tremendous effort has gone into developing protocols to infer hippocampal subfields from in vivo MRI. A fundamental challenge in this endeavor is that the cellular composition that defines the subfields currently cannot be captured with such techniques. In histology, there are typically sufficient microscale features available for an expert neuroanatomist to distinguish the subfields [1.Duvernoy H.M. et al.The Human Hippocampus: Functional Anatomy, Vascularization and Serial Sections with MRI. Springer Science & Business Media, 2013Crossref Scopus (111) Google Scholar,11.Ding S.-L. Van Hoesen G.W. Organization and detailed parcellation of human hippocampal head and body regions based on a combined analysis of cyto- and chemoarchitecture.J. Comp. Neurol. 2015; 523: 2233-2253Crossref PubMed Scopus (65) Google Scholar,12.Iglesias J.E. et al.A computational atlas of the hippocampal formation using ex vivo, ultra-high resolution MRI: application to adaptive segmentation of in vivo MRI.Neuroimage. 2015; 115: 117-137Crossref PubMed Scopus (535) Google Scholar], providing ground truth reference materials. Thus, the challenge of subfield segmentation is primarily in establishing correspondence between tissues and image modalities. The simplest method to do this is linear alignment between a given image and a ground truth 2D or 3D reference atlas, as in rigid stereotaxic approaches. However, this does not account for interindividual differences in hippocampal anatomy, which, as we describe in this opinion article, can vary widely. Currently, the issue of interindividual variability is addressed in one of three ways. First, manual segmentation aims to identify landmarks that remain consistently aligned to subfield boundaries despite interindividual differences. Second, alternatively, a reference atlas can be computationally deformed (or registered) to best fit a given subject, which is the workhorse of most automated segmentation methods. Third, novel surface-based methods aim to define hippocampal folding in 3D, which can then be mapped to a topologically constrained 2D space. In this opinion article, we argue that the latter approach holds unique promise because it is flexible to different variants of hippocampal folding and constrains alignment between individuals (or from one individual to a reference) in a biologically motivated manner. We see this approach as being aligned with the paradigm shift towards surface-based neocortical analysis methods that can accommodate different gyral and sulcal patterning [13.Van Essen D.C. et al.Functional and structural mapping of human cerebral cortex: solutions are in the surfaces.Adv. Neurol. 2000; 84: 23-34PubMed Google Scholar]. Hippocampal subfields are principally defined by cyto-, myelo-, and chemoarchitectures using ex vivo histology, typically in the coronal plane (Figure 1A ). When samples are taken from the body of the hippocampus (midway along the anterior–posterior extent), all subfields can be seen to follow a canonical topological ordering, which is shown at the top left of Figure 1A. Specifically, the subfields follow a curled C shape with dentate gyrus at the innermost terminus wrapped by cornu ammonis (CA) fields 4 through 1, followed by subiculum, which borders the medial temporal lobe neocortex. It has been known for a long time that, in 3D, the hippocampal subfields are fully contiguous [1.Duvernoy H.M. et al.The Human Hippocampus: Functional Anatomy, Vascularization and Serial Sections with MRI. Springer Science & Business Media, 2013Crossref Scopus (111) Google Scholar], even though this is often not apparent in individual (Figure this of the 3D of the hippocampus is through the of in S.-L. Van Hoesen G.W. Organization and detailed parcellation of human hippocampal head and body regions based on a combined analysis of cyto- and chemoarchitecture.J. Comp. Neurol. 2015; 523: 2233-2253Crossref PubMed Scopus (65) Google or the application of 3D imaging methods such as MRI. In all subfields follow a in The hippocampal and the hippocampal head and then where it on the The of the subfields in of regions has been in anatomical using coronal et provides of hippocampal 2020; PubMed Scopus Google Scholar], which we in Figure and in the the all in hippocampal that neocortical are seen to this can be most on coronal in the hippocampal head or on in the hippocampal as in Figure 1A. the of and the of anterior–posterior are to using histological methods which are typically to coronal have features apparent using 3D such as in vivo and ex vivo MRI. methods have to interindividual variability in the anterior–posterior of the hippocampus and in the and of (Figure S.-L. Van Hoesen G.W. Organization and detailed parcellation of human hippocampal head and body regions based on a combined analysis of cyto- and chemoarchitecture.J. Comp. Neurol. 2015; 523: 2233-2253Crossref PubMed Scopus (65) Google J. et the an for subfield and 2018; PubMed Scopus Google Scholar, et hippocampus with a analysis of the hippocampus on Scopus Google Scholar, J. in hippocampal 2020; Scholar]. that the of is linked to et hippocampus with a analysis of the hippocampus on Scopus Google and memory J. in hippocampal 2020; Scholar]. The of surface-based alignment is to account for interindividual differences in folding by to a 2D or or subfield segmentation in this unfolded can then be to the an approach that has been in the [13.Van Essen D.C. et al.Functional and structural mapping of human cerebral cortex: solutions are in the surfaces.Adv. Neurol. 2000; 84: 23-34PubMed Google and for In the a similar the and of the of the hippocampal along its anterior–posterior and its between or between one and a histological reference atlas, can then be by to this unfolded of in folding as in Figure of tissues between different individuals is in unfolded on the of and of in In surface-based approaches can accommodate pattern of and within the hippocampus and even such as hippocampal (or 3D et al.Hippocampal is an and has clinical in temporal lobe 2016; PubMed Scopus Google et abnormalities of the medial temporal lobe in with temporal lobe J. Google Scholar]. is to in have been that with such an parcellation of the methods for neocortical parcellation on surface-based which can be aligned in 2D that be to medial temporal lobe is the which and et al.A for manual segmentation of medial temporal lobe in Tesla Clin. 2017; PubMed Scopus Google of the in the human J. 2012; PubMed Scopus Google et of and from high-resolution the variability of the PubMed Scopus Google Scholar]. It is not 3D be applied to a with a or to a with a but a surface-based an to be fit to both The boundaries of or regions of by the in Figure can be applied in 2D on such a allowing to be topologically between individuals et method for and of and PubMed Scopus Google without or as topological which can a to or 2D neocortical are aligned to (e.g., the of the and major and features can be used to (e.g., et surface-based I. and 9: PubMed Scopus Google Scholar, et automated for the human cerebral on MRI into gyral based regions of PubMed Scopus Google Scholar, et and topologically of the human cerebral PubMed Scopus Google Scholar, et surface-based and a surface-based 9: PubMed Scopus Google Scholar], functional or other features et al.A parcellation of human cerebral 2016; PubMed Scopus Google of the hippocampal subfields as applied in J. et al.Hippocampal subfields through and of and morphological features in 3D 2020; PubMed Scopus Google data and are on hippocampal on the of where this borders in one in the top left in in in and lobe in to and by the as a allowing the hippocampus to be mapped to a 2D space. that the not in this unfolded due to its distinct J. et al.Hippocampal subfields through and of and morphological features in 3D 2020; PubMed Scopus Google for can be defined in this such as on the of 3D J. et al.Hippocampal subfields through and of and morphological features in 3D 2020; PubMed Scopus Google Scholar], and then can be to space. cornu dentate Figure image methods for neocortical parcellation on surface-based which can be aligned in 2D that be to medial temporal lobe is the which and et al.A for manual segmentation of medial temporal lobe in Tesla Clin. 2017; PubMed Scopus Google of the in the human J. 2012; PubMed Scopus Google et of and from high-resolution the variability of the PubMed Scopus Google Scholar]. It is not 3D be applied to a with a or to a with a but a surface-based an to be fit to both The boundaries of or regions of by the in Figure can be applied in 2D on such a allowing to be topologically between individuals et method for and of and PubMed Scopus Google without or as topological which can a to or 2D neocortical are aligned to (e.g., the of the and major and features can be used to (e.g., et surface-based I. and 9: PubMed Scopus Google Scholar, et automated for the human cerebral on MRI into gyral based regions of PubMed Scopus Google Scholar, et and topologically of the human cerebral PubMed Scopus Google Scholar, et surface-based and a surface-based 9: PubMed Scopus Google Scholar], functional or other features et al.A parcellation of human cerebral 2016; PubMed Scopus Google Scholar]). are on hippocampal on the of where this borders in one in the top left in in in and lobe in to and by the as a allowing the hippocampus to be mapped to a 2D space. that the not in this unfolded due to its distinct J. et al.Hippocampal subfields through and of and morphological features in 3D 2020; PubMed Scopus Google for can be defined in this such as on the of 3D J. et al.Hippocampal subfields through and of and morphological features in 3D 2020; PubMed Scopus Google Scholar], and then can be to space. cornu dentate have used surface-based methods to the hippocampus J. et the an for subfield and 2018; PubMed Scopus Google J. et al.Hippocampal subfields through and of and morphological features in 3D 2020; PubMed Scopus Google Scholar, et the human hippocampus with high resolution structural and functional PubMed Scopus Google Scholar, et in high-resolution imaging and computational of the human PubMed Scopus Google Scholar, R. et and of hippocampal subfield functional Sci. A. 2018; 115: PubMed Scopus Google Scholar]. In its and et the human hippocampus with high resolution structural and functional PubMed Scopus Google et in high-resolution imaging and computational of the human PubMed Scopus Google manual subfield segmentation in the of the with its subfield to an unfolded via methods similar to that have been used in the in human and the of functional PubMed Scopus Google et of in 2018; Scholar]. have similar of manual and surface-based but with mapping applied at the of individual which are then to a contiguous unfolded R. et and of hippocampal subfield functional Sci. A. 2018; 115: PubMed Scopus Google Scholar]. In J. et the an for subfield and 2018; PubMed Scopus Google J. et al.Hippocampal subfields through and of and morphological features in 3D 2020; PubMed Scopus Google Scholar], we have an approach that with of the hippocampal using a and then subfield segmentation is in unfolded (Figure of methods J. et the an for subfield and 2018; PubMed Scopus Google J. et al.Hippocampal subfields through and of and morphological features in 3D 2020; PubMed Scopus Google Scholar, et the human hippocampus with high resolution structural and functional PubMed Scopus Google Scholar, et in high-resolution imaging and computational of the human PubMed Scopus Google Scholar, R. et and of hippocampal subfield functional Sci. A. 2018; 115: PubMed Scopus Google allow measurement of detailed structural hippocampal such as or by of providing a that can be to the of hippocampal However, some of methods J. et the an for subfield and 2018; PubMed Scopus Google J. et al.Hippocampal subfields through and of and morphological features in 3D 2020; PubMed Scopus Google explicitly to to a in which reference subfield boundaries are The latter methods the topological of subfields and allow about subfield boundaries in 2D is by the of hippocampal which to within a archicortical sheet their folding [1.Duvernoy H.M. et al.The Human Hippocampus: Functional Anatomy, Vascularization and Serial Sections with MRI. Springer Science & Business Media, 2013Crossref Scopus (111) Google Scholar]. In manual protocols for subfield segmentation aim to provide for 2D coronal MRI to coronal via landmarks available in both modalities. Such the high and or et al.Hippocampal and memory in Alzheimer’s 2012; PubMed Scopus (111) Google between subfields in Figure 1, Figure as a in both containing and high and in MRI sufficient and it provides a to which subfield borders can be the between and be defined as of the of the et al.A for manual segmentation of medial temporal lobe in Tesla Clin. 2017; PubMed Scopus Google Scholar]. Thus, variability in the is to variability in the via manual protocols have been that the or other but show to an effort to this issue et of protocols for hippocampal subfields and in in vivo MRI: towards a segmentation 2015; PubMed Scopus Google Scholar, et al.A segmentation for hippocampal and do we one and are the 2017; 27: PubMed Scopus Google Scholar, R.K. et from the hippocampal subfields Google Scholar]. is that subfield boundaries do not remain in to the Specifically, this alignment is on the and along the anterior–posterior of the hippocampus S.-L. Van Hoesen G.W. Organization and detailed parcellation of human hippocampal head and body regions based on a combined analysis of cyto- and chemoarchitecture.J. Comp. Neurol. 2015; 523: 2233-2253Crossref PubMed Scopus (65) Google R. et of hippocampal subfields using ex vivo MRI and for in vivo 2020; PubMed Scopus Google Scholar]. because the hippocampal subfields (or of in most of the hippocampal head and their and can vary or between coronal can be by coronal with in Figure In there is interindividual variability in the and of within the hippocampal head S.-L. Van Hoesen G.W. Organization and detailed parcellation of human hippocampal head and body regions based on a combined analysis of cyto- and chemoarchitecture.J. Comp. Neurol. 2015; 523: 2233-2253Crossref PubMed Scopus (65) Google J. et al.Hippocampal subfields through and of and morphological features in 3D 2020; PubMed Scopus Google as as the hippocampal body and J. et the an for subfield and 2018; PubMed Scopus Google Scholar, et hippocampus with a analysis of the hippocampus on Scopus Google Scholar, J. in hippocampal 2020; (Figure the and it is not from the remain variants with different protocols or do not subfields in the head or because of et of protocols for hippocampal subfields and in in vivo MRI: towards a segmentation 2015; PubMed Scopus Google for approach to segmentation that is becoming increasingly computational methods. Such methods are because not are and but in can morphological 3D image features that be of plane to a or manual of MRI approaches of a given hippocampus to a detailed 3D reference which is via a of histology, ex vivo manual in vivo at high resolution J.E. et al.A computational atlas of the hippocampal formation using ex vivo, ultra-high resolution MRI: application to adaptive segmentation of in vivo MRI.Neuroimage. 2015; 115: 117-137Crossref PubMed Scopus (535) Google et al.A of manual and for hippocampal the 2018; PubMed Scopus Google et volumetry and analysis of hippocampal subfields and medial temporal in mild cognitive 2015; PubMed Scopus Google Scholar]. and both the and 3D reference used in methods J.E. et al.A computational atlas of the hippocampal formation using ex vivo, ultra-high resolution MRI: application to adaptive segmentation of in vivo MRI.Neuroimage. 2015; 115: 117-137Crossref PubMed Scopus (535) Google et al.A for manual segmentation of medial temporal lobe in Tesla Clin. 2017; PubMed Scopus Google et volumetry and analysis of hippocampal subfields and medial temporal in mild cognitive 2015; PubMed Scopus Google Scholar, et Hippocampal at J. 2016; 37: PubMed Scopus Google Scholar, J. et segmentation of the hippocampus and subfields using PubMed Scopus Google Scholar]. reference computational approaches can account for the seen in the hippocampal head and as as for subfield differences along the anterior–posterior of the hippocampal without challenges to or However, variability in hippocampal not be for registration-based approaches to it is not be applied to a hippocampus with to one with (Figure In some one reference or creating major and of anatomical in the 3D contiguous not the folding and topological in the is similar to the problem of the of individuals with or sulcal a problem that is in Currently, there is or to that surface-based subfield segmentation approaches other approaches in MRI with histological within coronal of the hippocampal and T.A. et of a histologically segmentation for the hippocampal 2017; PubMed Scopus Google histologically defined subfield boundaries as a along the of the hippocampus a topological and high and of topological more such as the being the of the which not remain the of the hippocampal body R. et of hippocampal subfields using ex vivo MRI and for in vivo 2020; PubMed Scopus Google Scholar]. a fully 3D approach (or 2D as in the surface-based approaches such are into the hippocampal head and However, there is of topologically defined subfield boundaries remain individuals in the hippocampal head and see J. et the an for subfield and 2018; PubMed Scopus Google Scholar]). differences in subfield boundaries for by folding or do and be as or boundaries in the unfolded in Figure in vivo MRI (e.g., or other that have been in neocortical et al.A parcellation of human cerebral 2016; PubMed Scopus Google to account for topologically differences in subfield locations between similar in to parcellation J. et al.Hippocampal subfields through and of and morphological features in 3D 2020; PubMed Scopus Google R. et and of hippocampal subfield functional Sci. A. 2018; 115: PubMed Scopus Google A. et and extensive the hippocampus PubMed Scopus (35) Google Scholar, et of and functional in the human 2020; PubMed Scopus Google Scholar, et to the of a Sci. 2018; 22: PubMed Scopus (78) Google Scholar]. histological that are ground it can be challenging to subfield boundaries between on such as or even different et al.A segmentation for hippocampal and do we one and are the 2017; 27: PubMed Scopus Google for be by using surface-based approaches to the 3D of a given or by 3D MRI Such in to ground truth subfield boundaries between within a surface-based see technical challenge with surface-based subfield segmentation approaches is that of different hippocampal is through detailed manual segmentation within the hippocampus J. et the an for subfield and 2018; PubMed Scopus Google J. et al.Hippocampal subfields through and of and morphological features in 3D 2020; PubMed Scopus Google Scholar]. However, or sclerosis (e.g., et al.Hippocampal internal architecture and in temporal lobe epilepsy due to hippocampal 2016; PubMed Scopus Google and with can the of the of this approach in (e.g., with or to in have been observed that with the of a surface-based even when applied at resolution or in with or Thus, despite limitations, it is opinion that surface-based hippocampal segmentation the in for interindividual differences when high-resolution data are which is becoming increasingly in the field at We have on the and folding within the human which challenges for manual and registration-based subfield segmentation methods in MRI. Specifically, manual approaches not the 3D of the registration-based approaches be to some of morphological the hippocampus as a folded and mapping it can account for this variability the problem of subfield segmentation from 3D to 2D and providing detailed morphological novel approach between hippocampal subfield integrity in disease, and of a surface-based subfield segmentation approach is and both a 3D model of hippocampal folding and histology, the hippocampal and 3D of hippocampal folding and mapping account for variability seen within subfield segmentation protocols often image but this can medial in the hippocampal head and and in the hippocampal is the of resolution and to a surface-based method to of the hippocampus can be through manual segmentation of and the which this be neocortical parcellation can be to accommodate features (e.g., or such approaches hippocampal surface-based subfield variants of hippocampal into distinct (e.g., three that into the hippocampal body and that vary in and the of that hippocampal is linked to aspects of memory and other functions linked to hippocampal folding more and their functional both from in such as or surface-based methods in such as or of hippocampal in or more conventional manual manual segmentation, or registration-based of a surface-based subfield segmentation approach is and both a 3D model of hippocampal folding and histology, the hippocampal and 3D of hippocampal folding and mapping account for variability seen within subfield segmentation protocols often image but this can medial in the hippocampal head and and in the hippocampal is the of resolution and to a surface-based method to of the hippocampus can be through manual segmentation of and the which this be neocortical parcellation can be to accommodate features (e.g., or such approaches hippocampal surface-based subfield variants of hippocampal into distinct (e.g., three that into the hippocampal body and that vary in and the of that hippocampal is linked to aspects of memory and other functions linked to hippocampal folding more and their functional both from in such as or surface-based methods in such as or of hippocampal in or more conventional manual manual segmentation, or registration-based We for on MRI methods and for on to The the of and and a The with through a fully 3D histological model of the and a 3D model by using 3D histological data from The 3D model on the left the and of the coronal and on the and of based on the used in Specifically, and a plane that best the in The 3D then along using linear for the subfield using

Ähnliche Arbeiten