Elsevier

NeuroImage

Volume 111, 1 May 2015, Pages 526-541
NeuroImage

Quantitative comparison of 21 protocols for labeling hippocampal subfields and parahippocampal subregions in in vivo MRI: Towards a harmonized segmentation protocol

https://doi.org/10.1016/j.neuroimage.2015.01.004Get rights and content

Highlights

  • We compare 21 manual protocols for labeling hippocampal and parahippocampal subfields.

  • 21 research groups applied their own manual segmentation protocol to the same anatomy.

  • Fuzzy similarity metrics used to quantify disagreement between protocols

  • Greatest disagreement is along the CA1/subiculum boundary, anterior hippocampus.

  • We propose a strategy for developing a harmonized segmentation protocol.

Abstract

Objective

An increasing number of human in vivo magnetic resonance imaging (MRI) studies have focused on examining the structure and function of the subfields of the hippocampal formation (the dentate gyrus, CA fields 1  3, and the subiculum) and subregions of the parahippocampal gyrus (entorhinal, perirhinal, and parahippocampal cortices). The ability to interpret the results of such studies and to relate them to each other would be improved if a common standard existed for labeling hippocampal subfields and parahippocampal subregions. Currently, research groups label different subsets of structures and use different rules, landmarks, and cues to define their anatomical extents. This paper characterizes, both qualitatively and quantitatively, the variability in the existing manual segmentation protocols for labeling hippocampal and parahippocampal substructures in MRI, with the goal of guiding subsequent work on developing a harmonized substructure segmentation protocol.

Method

MRI scans of a single healthy adult human subject were acquired both at 3 T and 7 T. Representatives from 21 research groups applied their respective manual segmentation protocols to the MRI modalities of their choice. The resulting set of 21 segmentations was analyzed in a common anatomical space to quantify similarity and identify areas of agreement.

Results

The differences between the 21 protocols include the region within which segmentation is performed, the set of anatomical labels used, and the extents of specific anatomical labels. The greatest overall disagreement among the protocols is at the CA1/subiculum boundary, and disagreement across all structures is greatest in the anterior portion of the hippocampal formation relative to the body and tail.

Conclusions

The combined examination of the 21 protocols in the same dataset suggests possible strategies towards developing a harmonized subfield segmentation protocol and facilitates comparison between published studies.

Introduction

The medial temporal lobe (MTL) is a complex brain region of enormous interest in research on memory, aging, psychiatric disorders, and neurodegenerative diseases. Within the MTL, the subfields of the hippocampus (cornu Ammonis fields CA1  CA4, dentate gyrus, subiculum) and the adjacent cortical subregions of the parahippocampal gyrus (entorhinal cortex, perirhinal cortex, and parahippocampal cortex) are understood to subserve different functions in the memory system (Squire et al., 2004, Moscovitch et al., 2006, Bakker et al., 2008, Wolk et al., 2011). Different psychiatric and neurological disorders are known to affect hippocampal subfields and MTL cortical subregions differently, selectively, and in a complex progression (Braak and Braak, 1995, Arnold et al., 1995, Simić et al., 1997, de Lanerolle et al., 2003, West et al., 2004, Lucassen et al., 2006, Small et al., 2011). The non-uniformity of MTL involvement in normal brain function and in disease makes in vivo interrogation of the structural and functional properties of hippocampal subfields and parahippocampal subregions highly desirable. Recent advances in MRI technology have made it possible to visualize the hippocampal region with increasing detail, leading a growing number of researchers to attempt to label and quantify small substructures using in vivo MRI (Insausti et al., 1998, Small et al., 2000, Zeineh et al., 2001, Zeineh et al., 2003, Zeineh et al., 2012, Wang et al., 2003, Wang et al., 2006, Wang et al., 2010, Apostolova et al., 2006, Kirwan et al., 2007, Mueller et al., 2007, Mueller and Weiner, 2009, Van Leemput et al., 2009, Ekstrom et al., 2009, Fischl et al., 2009, Olsen et al., 2009, Olsen et al., 2013, Malykhin et al., 2010, Kerchner et al., 2010, Kerchner et al., 2012, Preston et al., 2010, Prudent et al., 2010, Yassa et al., 2010, La Joie et al., 2010, La Joie et al., 2013, Hanseeuw et al., 2011, Henry et al., 2011, Bonnici et al., 2012, Wisse et al., 2012, Pluta et al., 2012, Teicher et al., 2012, Libby et al., 2012, Bender et al., 2013, Winterburn et al., 2013, Kirov et al., 2013, Augustinack et al., 2013, Palombo et al., 2013, Pereira et al., 2013, Duncan et al., 2014, Yushkevich et al., 2015).

However, the anatomy of the human MTL is complex and variable, and the boundaries between different subfields have been described in the neuroanatomy literature using cytoarchitectonic features that require histological staining and microscopic resolution to visualize (Lorente de Nó, 1934, Rosene and Van Hoesen, 1987, Gloor, 1997, Insausti and Amaral, 2004, Duvernoy, 2005, Amaral and Lavenex, 2007, van Strien et al., 2012). Even at that resolution, neuroanatomical references do not always agree on the definition and boundaries of subfields. Any protocol that attempts to label these substructures in MRI, regardless of resolution, has to employ some combination of image intensity cues, known anatomical landmarks, and geometrical rules to define boundaries between substructures. A substantial number of manual segmentation protocols have been published in the last few years, and up to now, no common set of rules has been adopted by the research community. Indeed, different groups partition the MTL into different subsets of substructures, with different rules used to define each substructure, and different extents of the region within which the substructures are labeled. For example, one protocol may combine all CA subfields into a single label, draw the boundary between CA1 and subiculum at the medial-most extent of the dentate gyrus, and exclude the hippocampal head and tail from the segmentation. Another protocol may group CA3 and the dentate gyrus into one label and draw the CA1/subiculum boundary in a more lateral location, while also labeling the full extent of the hippocampus. Such variability among protocols makes comparisons between the results reported by different research groups difficult.

In this paper, we take the first step towards quantitatively and qualitatively characterizing the differences between the hippocampal subfield and parahippocampal subregion segmentation protocols used in the in vivo imaging community. We do so by having 21 research groups apply their manual segmentation protocols to label the left MTL of the same subject, which makes it possible for the segmentations to be compared on a voxel by voxel basis. Since different groups have used different MRI field strengths and different MRI contrast mechanisms to develop their protocols, the single subject in this study was scanned using three different MRI protocols (T1-weighted 3 T MRI, T2-weighted 3 T MRI, and T2-weighted 7 T MRI), and participating research groups chose the images that best fitted the MRI modality targeted by their respective protocols. We report on the differences in label sets used by the different protocols, provide voxel-wise maps of inter-protocol agreement, and identify substructure boundaries where there is most disagreement between protocols.

This work follows in the footsteps of an analogous investigation of whole hippocampus segmentation protocols carried out by the EADC-ADNI work group (Boccardi et al., 2011), with several important distinctions. In the EADC-ADNI effort, the hippocampus was labeled as a single structure; the segmentations were performed centrally by a single rater and subsequently checked and certified by the protocols' authors; and the comparisons were carried out at a qualitative level. In contrast, the present study addresses a more complex neuroanatomical problem with a large number of substructures, and performs quantitative comparisons on manual segmentations provided by the protocol developers themselves in different MRI modalities. Moreover, whereas the EADC-ADNI effort performed their comparison using 12 representative protocols from a much larger number of available whole-hippocampus MRI segmentation protocols, our study is able to include most of the published protocols for hippocampal/parahippocampal subfield segmentation in MRI. This broad inclusion is made possible by the smaller size of the subfield neuroimaging research community, but also by our decision not to restrict the comparison to a single MRI field strength or modality.

The EADC-ADNI work group successfully used the protocol comparison in (Boccardi et al., 2011) as the first step towards reconciling differences among those protocols, which in turn led to the development of a highly reliable harmonized whole hippocampus segmentation protocol (Boccardi et al., 2013, Boccardi et al., 2014, Bocchetta et al., 2014). Inspired by the success of the EADC-ADNI effort, we similarly envision the quantitative characterization of the differences and commonalities across the 21 protocols in this study becoming the first step towards developing a unified, harmonized subfield segmentation protocol.

Section snippets

Magnetic resonance imaging

MRI scans from one 36 year old male right-handed subject with no history of neurologic or psychiatric disease were analyzed in this study. Scans were acquired as part of an MRI technology development protocol at the University of Pennsylvania. Informed consent was obtained in accordance with the University of Pennsylvania Institutional Review Board (IRB).

The subject was first scanned on the Siemens Trio 3 Tesla MRI scanner using a 32 channel head receiver array. The protocol included a

Qualitative Comparison

Fig. 2, Fig. 3 show the 21 segmentations resampled into the common image space at oblique coronal slices through the hippocampal head and body.3 Each group's segmentation is superimposed on the MRI modality used by that group. Additionally, Fig. 4 shows the 3D renderings of the 21 segmentations in the common space. The figures make it possible to compare segmentation protocols side by side

Discussion

This is the first study to directly examine agreement between a large number of hippocampal subfield and parahippocampal cortical subregion segmentation protocols in a common image dataset. The study reveals significant variability among the protocols currently used in the field in terms of what labels are used, where the boundaries between labels are placed, and what extent of the hippocampal region is labeled. Nonetheless, by quantifying this variability and identifying regions of greatest

Conclusions

This study has for the first time compared a large number of protocols for the segmentation of hippocampal subfields and parahippocampal subregions in a common MRI dataset. The comparison demonstrates the challenges facing future efforts towards protocol harmonization. Existing protocols vary in the sets of labels used, the rules used to define subfield boundaries, the anterior–posterior extents of the segmentation, the sources and the purposes of the protocols. These differences limit the

Acknowledgment

Many of the 21 segmentation protocols are the result of extensive development effort by many contributors who could not be included as authors, and we graciously acknowledge their contributions to this work. We thank the participants of the Hippocampal Subfield Segmentation Summit (HS3) series of meetings in 2013–14 for their insightful comments, which have influenced this paper extensively. In particular, we especially thank Prof. Ricardo Insausti and Dr. Niels van Strien for their support of

References (81)

  • I.I. Kirov et al.

    In vivo 7 tesla imaging of the dentate granule cell layer in schizophrenia

    Schizophr. Res.

    (2013)
  • R. La Joie et al.

    Differential effect of age on hippocampal subfields assessed using a new high-resolution 3T MR sequence

    Neuroimage

    (2010)
  • R. La Joie et al.

    Hippocampal subfield volumetry in mild cognitive impairment, Alzheimer's disease and semantic dementia

    NeuroImage

    (2013)
  • N.V. Malykhin et al.

    Three-dimensional volumetric analysis and reconstruction of amygdala and hippocampal head, body and tail

    Psychiatry Res.

    (2007)
  • N.V. Malykhin et al.

    In vivo quantification of hippocampal subfields using 4.7 T fast spin echo imaging

    Neuroimage

    (2010)
  • M. Moscovitch et al.

    The cognitive neuroscience of remote episodic, semantic and spatial memory

    Curr. Opin. Neurobiol.

    (2006)
  • S.G. Mueller et al.

    Measurement of hippocampal subfields and age-related changes with high resolution MRI at 4 T

    Neurobiol. Aging

    (2007)
  • J. Pipitone et al.

    Multi-atlas segmentation of the whole hippocampus and subfields using multiple automatically generated templates

    Neuroimage

    (2014)
  • L. Wang et al.

    Changes in hippocampal volume and shape across time distinguish dementia of the Alzheimer type from healthy aging

    Neuroimage

    (2003)
  • L. Wang et al.

    Abnormalities of hippocampal surface structure in very mild dementia of the Alzheimer type

    Neuroimage

    (2006)
  • M.J. West et al.

    Hippocampal neurons in pre-clinical Alzheimer's disease

    Neurobiol. Aging

    (2004)
  • J.L. Winterburn et al.

    A novel in vivo atlas of human hippocampal subfields using high-resolution 3 t magnetic resonance imaging

    Neuroimage

    (2013)
  • L.E.M. Wisse et al.

    Subfields of the hippocampal formation at 7 T MRI: in vivo volumetric assessment

    Neuroimage

    (2012)
  • M.A. Yassa et al.

    High-resolution structural and functional MRI of hippocampal CA3 and dentate gyrus in patients with amnestic mild cognitive impairment

    Neuroimage

    (2010)
  • M.M. Zeineh et al.

    Application of cortical unfolding techniques to functional MRI of the human hippocampal region

    Neuroimage

    (2000)
  • M.M. Zeineh et al.

    Ultra-high resolution diffusion tensor imaging of the microscopic pathways of the medial temporal lobe

    Neuroimage

    (2012)
  • D.G. Amaral et al.

    The hippocampal formation

  • D. Amaral et al.

    Hippocampal Neuroanatomy

  • L.G. Apostolova et al.

    3D comparison of hippocampal atrophy in amnestic mild cognitive impairment and Alzheimer's disease

    Brain

    (2006)
  • S.E. Arnold et al.

    Smaller neuron size in schizophrenia in hippocampal subfields that mediate cortical–hippocampal interactions

    Am. J. Psychiatry

    (1995)
  • A. Bakker et al.

    Pattern separation in the human hippocampal CA3 and dentate gyrus

    Science

    (2008)
  • A.R. Bender et al.

    Vascular risk moderates associations between hippocampal subfield volumes and memory

    J. Cogn. Neurosci.

    (2013)
  • M. Boccardi et al.

    Survey of protocols for the manual segmentation of the hippocampus: preparatory steps towards a joint EADC-ADNI harmonized protocol

    J. Alzheimers Dis.

    (2011)
  • M. Boccardi et al.

    Operationalizing protocol differences for EADC-ADNI manual hippocampal segmentation

    Alzheimers Dement.

    (2013)
  • M. Boccardi et al.

    Delphi definition of the EADC-ADNI harmonized protocol for hippocampal segmentation on magnetic resonance

    Alzheimers Dement.

    (2014)
  • M. Bocchetta et al.

    Harmonized benchmark labels of the hippocampus on magnetic resonance: the EADC-ADNI project

    Alzheimers Dement.

    (2014)
  • H.M. Bonnici et al.

    Multi-voxel pattern analysis in human hippocampal subfields

    Front. Hum. Neurosci.

    (2012)
  • N.C. de Lanerolle et al.

    A retrospective analysis of hippocampal pathology in human temporal lobe epilepsy: evidence for distinctive patient subcategories

    Epilepsia

    (2003)
  • L.R. Dice

    Measures of the amount of ecologic association between species

    Ecology

    (1945)
  • K. Duncan et al.

    Associative encoding and retrieval are predicted by functional connectivity in distinct hippocampal area ca1 pathways

    J. Neurosci.

    (2014)
  • Cited by (238)

    View all citing articles on Scopus
    View full text