Automated segmentation of hippocampal subfields from ultra‐high resolution in vivo MRI
Open accessAbstract
Recent developments in MRI data acquisition technology are starting to yield images that show anatomical features of the hippocampal formation at an unprecedented level of detail, providing the basis for hippocampal subfield measurement. However, a fundamental bottleneck in MRI studies of the hippocampus at the subfield level is that they currently depend on manual segmentation, a laborious process that severely limits the amount of data that can be analyzed. In this article, we present a computational method for segmenting the hippocampal subfields in ultra‐high resolution MRI data in a fully automated fashion. Using Bayesian inference, we use a statistical model of image formation around the hippocampal area to obtain automated segmentations. We validate the proposed technique by comparing its segmentations to corresponding manual delineations in ultra‐high resolution MRI scans of 10 individuals, and show that automated volume measurements of the larger subfields correlate well with manual volume estimates. Unlike manual segmentations, our automated technique is fully reproducible, and fast enough to enable routine analysis of the hippocampal subfields in large imaging studies. © 2009 Wiley‐Liss, Inc.
AI snapshot
Written from the paper’s full text.
TL;DR
The paper presents a fully automated Bayesian method for segmenting hippocampal subfields from ultra-high resolution in vivo MRI, using a mesh-based probabilistic atlas learned from manual delineations. In 10 subjects, automated volumes of larger subfields correlated well with manual estimates, while the method was fully reproducible and took about 5 h per subject.
Key findings
- Automated segmentation took about 5 h per subject on a 2.83 GHz Intel Xeon E5440 processor, whereas manual labeling took approximately 2.5 days per hippocampus.
- Dice overlap was highest for larger structures: CA2/3 and subiculum (average sizes 935 and 537 mm³) had average Dice around 0.74; CA4/DG (526 mm³) and presubiculum (431 mm³) scored around 0.68 each; CA1 (340 mm³) scored around 0.62; fimbria and hippocampal fissure (81 mm³ each) scored around 0.51 and 0.53.
- Automated volumes correlated with manual volumes for CA2/3 (r = 0.91, P ≤ 0.0002) and CA4/DG (r = 0.83, P ≤ 0.0028); subiculum correlated at r = 0.60 (P ≤ 0.066), and hippocampal fissure at r = 0.50 (P ≤ 0.14), with systematic underestimation of fissure volume.
- Average boundary error was below the 380 µm voxel size for most structures, including fimbria; exceptions were CA1 (about 1.17 times voxel size) and hippocampal fissure (almost twice voxel size). Mean intrarater Dice overlap over all structures was 0.79, and automated results were systematically more different from manual segmentations than repeated manual segmentations were from one another.
Methods
- Design: Bayesian generative model of MRI image formation around the hippocampal area, combining a mesh-based probabilistic atlas prior with a Gaussian likelihood model; parameters optimized via Generalized Expectation-Maximization and MAP estimation, with multi-resolution optimization.
- Data/sample: Ultra-high resolution T1-weighted MPRAGE scans at 3.0T (380 µm in-plane, TR/TI/TE = 2530/1100/5.39 ms, FOV = 448, FA = 7°, 208 coronal slices, thickness 0.8 mm, acquisition time 7.34 min, five acquisitions) from 10 subjects (six younger, four older cognitively normal; five male, five female; age range 22–89); right hippocampus manually delineated.
- Analysis: Leave-one-out cross-validation, building an atlas from nine subjects and segmenting the held-out subject; compared automated versus manual segmentations using Dice overlap, average boundary distance, Pearson correlation of volumes, and comparison to repeated manual delineations in two midbody slices from five subjects.
Limitations
- Stated: Manual segmentations were treated as gold standard but suffer their own errors, including greater consistency in the plane drawn than in cross-sectional directions and possible repeated mistakes by the same operator; a more comprehensive validation would require multiple human operators and repeat scans, which was outside the scope.
- Stated: The manual delineation protocol was still being validated and refined and was used in its initial form; the caudal “catch-all” hippocampus label varied from 5 to 17% of total hippocampal volume and was excluded from comparisons because its starting point was unavailable to the automated algorithm.
- Stated/evident: Results are based on T1-weighted images only, with future work needed for T2-weighted or multi-spectral data; the validation sample is small (10 subjects, right hippocampus only) and uses a single-site, single-operator manual reference.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Claims the paper makes
0 of 4 cited sources checked · 0 supported · 0 flagged
- Findingown data
Manual delineation of hippocampal subfields in ultra-high resolution images is extremely time-consuming, taking several days for a full hippocampus in the authors' images.
“delineating subfields manually in ultrahigh resolution images is an excruciatingly time consuming process (several days for a full hippocampus in our images).”
- Also stated in: § EXPERIMENTS · plain
“Using this procedure, each hippocampus takes approximately 2.5 days to label.”
Quote in the paperSubstantial claim - Also stated in: § EXPERIMENTS · plain
- Findingown data
Automated volume measurements of the larger hippocampal subfields correlate well with manual volume estimates.
“show that automated volume measurements of the larger subfields correlate well with manual volume estimates.”
- Also stated in: § DISCUSSION · plain
“We showed that automated volume measurements of the larger subfields CA2/3, CA4/DG, and, to a lesser degree, subiculum, correlated well with manual volume estimates.”
Quote in the paperSubstantial claim- The authors state this as an assumption: Manual delineations provide a valid reference standard ('gold standard') for evaluating the automated segmentation algorithm.
- Also stated in: § DISCUSSION · plain
- Findingown data
CA2/3 and subiculum had average Dice coefficients of around 0.74 between automated and manual segmentation.
“CA2/3 and subiculum, the largest structures with an average size of 935 and 537 mm3, respectively, have an average Dice coefficient of around 0.74”
Quote in the paperSubstantial claim- The authors state this as an assumption: Manual delineations provide a valid reference standard ('gold standard') for evaluating the automated segmentation algorithm.
- Findingown data
CA4/DG and presubiculum had average Dice coefficients of around 0.68 between automated and manual segmentation.
“whereas CA4/DG (on average 526 mm3) and presubiculum (on average 431 mm3) score around 0.68 each”
Quote in the paperSubstantial claim- The authors state this as an assumption: Manual delineations provide a valid reference standard ('gold standard') for evaluating the automated segmentation algorithm.
- Findingown data
Automated volumes of CA2/3 are strongly correlated with manual volumes (correlation coefficient 0.91, P ≤ 0.0002).
“the automatically calculated volumes of CA2/3 and CA4/DG are strongly correlated with the manual ones, with a correlation coefficient of 0.91 (P ≤ 0.0002) and 0.83 (P ≤ 0.0028), respectively.”
Quote in the paperSubstantial claim- The authors state this as an assumption: Manual delineations provide a valid reference standard ('gold standard') for evaluating the automated segmentation algorithm.
- Findingown data
Automated volumes of CA4/DG are strongly correlated with manual volumes (correlation coefficient 0.83, P ≤ 0.0028).
“the automatically calculated volumes of CA2/3 and CA4/DG are strongly correlated with the manual ones, with a correlation coefficient of 0.91 (P ≤ 0.0002) and 0.83 (P ≤ 0.0028), respectively.”
Quote in the paperSubstantial claim- The authors state this as an assumption: Manual delineations provide a valid reference standard ('gold standard') for evaluating the automated segmentation algorithm.
- Findingown data
CA1 has an average boundary error of about 1.17 times the voxel size.
“Exceptions are CA1, which has an average boundary error of about 1.17 times the voxel size”
Quote in the paperNot a substantial claim- The authors state this as an assumption: Manual delineations provide a valid reference standard ('gold standard') for evaluating the automated segmentation algorithm.
- Findingown data
Manual delineations typically appear more consistent in the plane they were drawn in than in other, cross-sectional directions.
“manual delineations typically appear more consistent in the plane they were drawn in than in other, cross-sectional directions, as can be seen in Figure 1.”
Quote in the paperNot a substantial claim - Findingown data
Larger hippocampal structures score better on Dice overlap between automated and manual segmentation than smaller ones.
“As expected, larger structures score better than do smaller ones: CA2/3 and subiculum, the largest structures with an average size of 935 and 537 mm3, respectively, have an average Dice coefficient of around 0.74, whereas CA4/DG (on average 526 mm3) and presubiculum (on average 431 mm3) score around 0.68 each, and the smaller CA1 (on average 340 mm3) has a Dice coefficient of around 0.62.”
Quote in the paperNot a substantial claim- The authors state this as an assumption: Manual delineations provide a valid reference standard ('gold standard') for evaluating the automated segmentation algorithm.
- Findingown data
The average boundary error between automated and manual segmentation is below the voxel size of 380 μm for most structures, including the fimbria.
“The average boundary error is below the voxel size of 380 µm for most structures, including the fimbria.”
Quote in the paperSubstantial claim- The authors state this as an assumption: Manual delineations provide a valid reference standard ('gold standard') for evaluating the automated segmentation algorithm.
- Findingown data
The hippocampal fissure has an average boundary error almost twice the voxel size.
“and the hippocampal fissure, with an average error almost twice the voxel size.”
Quote in the paperNot a substantial claim- The authors state this as an assumption: Manual delineations provide a valid reference standard ('gold standard') for evaluating the automated segmentation algorithm.
- Findingown data
Mean intrarater Dice overlap over all structures was 0.79.
“mean intrarater Dice overlap over all structures: 0.79”
Quote in the paperSubstantial claim- The authors state this as an assumption: Manual delineations provide a valid reference standard ('gold standard') for evaluating the automated segmentation algorithm.
- Findingown data
CA1 had an average Dice coefficient of around 0.62 between automated and manual segmentation.
“and the smaller CA1 (on average 340 mm3) has a Dice coefficient of around 0.62.”
Quote in the paperSubstantial claim- The authors state this as an assumption: Manual delineations provide a valid reference standard ('gold standard') for evaluating the automated segmentation algorithm.
- Findingown data
The automated results are systematically more different from the manual segmentations than the manual segmentations are from one another.
“With respect to the intrarater variability of the human operator, the automated results are systematically more different from the manual segmentations than the manual segmentations from one another”
Quote in the paperSubstantial claim- The authors state this as an assumption: Manual delineations provide a valid reference standard ('gold standard') for evaluating the automated segmentation algorithm.
- Findingown data
Automated volumes of the hippocampal fissure correlate with manual volumes (correlation coefficient 0.50, P ≤ 0.14).
“its automated volume measurements correlate better with the manual ones than do some structures with much better segmentation evaluation metrics (correlation coefficient 0.50, P ≤ 0.14).”
Quote in the paperNot a substantial claim- The authors state this as an assumption: Manual delineations provide a valid reference standard ('gold standard') for evaluating the automated segmentation algorithm.
- Premisereasoning
A fundamental bottleneck in MRI studies of the hippocampus at the subfield level is that they currently depend on manual segmentation, which severely limits the amount of data that can be analyzed.
“However, a fundamental bottleneck in MRI studies of the hippocampus at the subfield level is that they currently depend on manual segmentation, a laborious process that severely limits the amount of data that can be analyzed.”
- Also stated in: § INTRODUCTION · plain
“While the high field MRI technology needed for visualizing hippocampal subfields is becoming increasingly available, a fundamental bottleneck in current imaging studies is that manual interaction is required to draw subfield boundaries in the images. This severely limits the amount of data that can be analyzed”
Quote in the paperSubstantial claim - Also stated in: § INTRODUCTION · plain
- Interpretationreasoning
Manual measurements suffer from their own type of errors.
“it should be noted that manual measurements suffer from their own type of errors as well.”
Quote in the paperNot a substantial claim - Interpretationown data
The relatively poor segmentation evaluation scores for the hippocampal fissure are caused by a systematic underestimation of its volume by the automated method.
“The relatively poor segmentation evaluation scores are apparently caused by a systematic underestimation of the volume of the hippocampal fissure by the automated method.”
Quote in the paperNot a substantial claim- The authors state this as an assumption: Manual delineations provide a valid reference standard ('gold standard') for evaluating the automated segmentation algorithm.
- Interpretationown data
The automated segmentation technique is fast enough to enable routine analysis of the hippocampal subfields in large imaging studies.
“and fast enough to enable routine analysis of the hippocampal subfields in large imaging studies.”
- Also stated in: § DISCUSSION · plain
“and fast enough to enable routine analysis of the hippocampal subfields in large imaging studies.”
Quote in the paperNot a substantial claim - Also stated in: § DISCUSSION · plain
- Study design and conductreasoning
The GEM algorithm guarantees that the objective function is improved with each new iteration, unless a local maximum or saddle point has already been reached.
“Since the lower bound always touches the objective function at the current parameter estimate, this procedure guarantees that the objective function is improved with each new iteration, unless a local maximum or saddle point has already been reached.”
Quote in the paperNot a substantial claim - Interpretationnone given
The automated segmentation technique is fully reproducible.
“Unlike manual segmentations, our automated technique is fully reproducible”
- Also stated in: § DISCUSSION · plain
“Unlike manual segmentations, our automated technique is fully reproducible”
Quote in the paperNot a substantial claim - Also stated in: § DISCUSSION · plain
- May combine two claims. 8
- Classified as probably not a substantial claim. 8
May combine two claims.
Findingown dataFimbria and hippocampal fissure had average Dice coefficients of around 0.51 and 0.53, respectively, between automated and manual segmentation.
“Automated segmentation of the fimbria and the hippocampal fissure, the smallest structures with an average volume of only 81 mm3 each, is considerably more challenging, with a Dice coefficient of around 0.51 and 0.53, respectively.”
Quote in the paperSubstantial claim- The authors state this as an assumption: Manual delineations provide a valid reference standard ('gold standard') for evaluating the automated segmentation algorithm.
May combine two claims.
Findingown dataThe automated method performs similarly on selected midbody slices as on whole volumes, except for the presubiculum (Dice coefficient 0.55) and the hippocampal fissure (Dice coefficient 0.2).
“Qualitatively, the automated method performs similarly on these selected slices as on the whole volumes, except for the presubiculum (Dice coefficient decreased to 0.55), and the hippocampal fissure (Dice coefficient only 0.2).”
Quote in the paperSubstantial claim- The authors state this as an assumption: Manual delineations provide a valid reference standard ('gold standard') for evaluating the automated segmentation algorithm.
Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.
Findingown dataThe computational segmentation process for each of the 10 subjects was fully automated and took about 5 h per subject on a 2.83 GHz Intel Xeon E5440 processor.
“The computational segmentation process for each of the 10 subjects was fully automated and took about 5 h per subject on a 2.83 GHz Intel Xeon E5440 processor.”
Quote in the paperNot a substantial claimMay combine two claims.
Findingown dataAutomated volumes of subiculum correlate with manual volumes (correlation coefficient 0.60) but the correlation does not reach statistical significance (P ≤ 0.066).
“Subiculum also correlates to some degree (correlation coefficient 0.60), although the correlation does not reach statistical significance (P ≤ 0.066).”
Quote in the paperNot a substantial claim- The authors state this as an assumption: Manual delineations provide a valid reference standard ('gold standard') for evaluating the automated segmentation algorithm.
May combine two claims.
Findingown dataWith the exception of the hippocampal fissure, the structures most difficult to segment reliably by the human operator are also the most difficult for the automated method; the order in which Dice scores decrease across structures is the same for both methods.
“With the exception of the hippocampal fissure, the structures that are most difficult to segment reliably by the human operator are also the most difficult for the automated method: the order in which the Dice scores decrease across structures is the same for both methods.”
Quote in the paperNot a substantial claim- The authors state this as an assumption: Manual delineations provide a valid reference standard ('gold standard') for evaluating the automated segmentation algorithm.
Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).
Interpretationnone givenGeneralizing the model and optimization procedures from single-valued to vector-valued intensities is straightforward, allowing the algorithm to directly analyze multi-spectral MRI scans.
“Furthermore, generalizing the model and the optimization procedures described here from single-valued to vector-valued intensities is straightforward, allowing the algorithm to directly analyze multi-spectral MRI scans as well.”
Quote in the paperNot a substantial claimClassified as probably not a substantial claim.
Study design and conductown dataTo place automated performance in the context of human intrarater variability, two consecutive coronal slices in the midbody of the hippocampus in five subjects were re-delineated twice by the same human operator.
“To place the automated method’s segmentation performance in the context of human intrarater variability, two consecutive coronal slices in the midbody of the hippocampus in a randomly chosen subset of five subjects (three younger and two older subjects; two male and three female; age range 22–89) were re-delineated twice by the same human operator who performed the original segmentations.”
Quote in the paperSubstantial claim- The authors state this as an assumption: Manual delineations provide a valid reference standard ('gold standard') for evaluating the automated segmentation algorithm.
Classified as probably not a substantial claim.
Study design and conductown dataThe automated segmentation algorithm was evaluated using a leave-one-out cross-validation strategy: an atlas mesh was built from manual delineations in nine of the 10 subjects and used to segment the remaining subject, repeated for each subject.
“We evaluated our automated segmentation algorithm using a leave-one-out cross-validation strategy: we built an atlas mesh from the manual delineations in nine of the 10 subjects, and used this to segment the image of the remaining subject. We repeated this process for each of the 10 subjects”
Quote in the paperSubstantial claim- The authors state this as an assumption: Manual delineations provide a valid reference standard ('gold standard') for evaluating the automated segmentation algorithm.
Probably not a substantial claim: may not pass S3 (load-bearing).
InterpretationreasoningThe algorithm automatically learns and adapts to each scan's contrast properties, which should make it applicable to images acquired with different pulse sequences, such as T2-weighted protocols.
“While the results presented here are based on T1-weighted images, the fact that our algorithm automatically learns and adapts to each scan’s contrast properties should make it applicable to images acquired with different pulse sequences as well, for example, the T2-weighted protocols used by some other groups”
Quote in the paperNot a substantial claimMay combine two claims.
PremisecitationThe distinct subregions composing the hippocampus have been implicated in different memory subsystems and shown to be differentially affected in normal aging and AD.
“The distinct subregions composing the hippocampus have been implicated in different memory subsystems (Gabrieli et al., 1997; Zeineh et al., 2003), and shown to be differentially affected in normal aging and AD (West et al., 1994; Fukutani et al., 1995; Small et al., 1999; Small et al., 2000).”
Quote in the paperSubstantial claim0 of 3 sources checked- cites Gabrieli et al., 1997
Not checked in the free run.
- cites Zeineh et al., 2003
Not checked in the free run.
- cites West et al., 1994
Not checked in the free run.
Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.
Study design and conductnone givenThe manual delineation protocol used for reference segmentations is currently being validated and refined and was used in its initial form.
“The delineation protocol is currently being validated and refined and is used here in its initial form for the primary purpose of the development and reliability assessment of the automated segmentation technique that is the main focus of the present study.”
Quote in the paperNot a substantial claim- Rests on an assumption: The manual delineation protocol, though still being validated and refined, is adequate for the primary purpose of development and reliability assessment of the automated segmentation technique.
May combine two claims.
PremisecitationThe probabilistic atlas learning procedure automatically yields sparse atlas representations that avoid overfitting to training data and are better at predicting neuroanatomy in new subjects than conventional probabilistic atlases.
“As shown in (Van Leemput, (in press)), this procedure automatically yields sparse atlas representations that explicitly avoid overfitting to the training data, and that are therefore better at predicting the neuroanatomy in new subjects than conventional probabilistic atlases.”
Quote in the paperSubstantial claim0 of 1 sources checked- cites Van Leemput, (in press)
Not checked in the free run.
Classified as probably not a substantial claim.
Study design and conductreasoningVoxels labeled as the 'catch-all' hippocampus label in either the automated or manual segmentation were excluded from comparisons.
“voxels that were labeled as such in either the automated or manual segmentation were not included in the comparisons.”
Quote in the paperSubstantial claim- The authors state this as an assumption: Manual delineations provide a valid reference standard ('gold standard') for evaluating the automated segmentation algorithm.
Classified as probably not a substantial claim.
Study design and conductreasoningThe study used a multi-resolution optimization strategy to reduce the risk of getting trapped in a local optimum.
“In order to reduce the risk of getting trapped in a local optimum, we employ a multi-resolution optimization strategy”
Quote in the paperSubstantial claim
The authors state this as an assumption.
Manual delineations provide a valid reference standard ('gold standard') for evaluating the automated segmentation algorithm.
“Although manual segmentations were considered the gold standard for evaluating our automated algorithm”
The paper relies on this without support.
The manual delineation protocol, though still being validated and refined, is adequate for the primary purpose of development and reliability assessment of the automated segmentation technique.
“The delineation protocol is currently being validated and refined and is used here in its initial form for the primary purpose of the development and reliability assessment of the automated segmentation technique that is the main focus of the present study.”
The paper relies on this without support.
Manual delineations suffer from intra- and interobserver variability, which confounds subsequent statistical analyses of the results.
“Furthermore, manual delineations suffer from intra- and interobserver variability, which confounds subsequent statistical analyses of the results.”
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
The Dice overlap measures between manual and automated segmentation methods were approximately 0.7 for all the substructures.
“The Dice overlap measures between manual and automated segmentation methods were approximately 0.7 for all the substructures [28].”
From Atrophy of presubiculum and subiculum is the earliest hippocampal anatomical marker of Alzheimer's disease (2015) · cites this paper as [28]
The original MRI sequence used by van Leemput et al. was about 35 minutes long, making it useless in clinical settings, especially for degenerative diseases, due to unfitness for patients and proneness to movement artifacts.
“Although we are aware that this step could introduce some biases, it should be outlined that the original MRI sequence used by van Leemput et al. [28] was about 35 minutes long. This makes it useless in clinical settings, especially in the framework of degenerative diseases, given its unfitness for patients and its proneness to movement artifacts.”
From Atrophy of presubiculum and subiculum is the earliest hippocampal anatomical marker of Alzheimer's disease (2015) · cites this paper as [28]
The difference in resolution between the present study's images (1×1×1 mm) and those originally used by van Leemput et al. (0.38×0.38×0.8) is a critical aspect.
“First, a critical aspect is the difference in resolution between our images (1 × 1 × 1 mm) and those originally used by van Leemput et al. [28] (0.38 × 0.38 × 0.8).”
From Atrophy of presubiculum and subiculum is the earliest hippocampal anatomical marker of Alzheimer's disease (2015) · cites this paper as [28]
Hippocampal subfield volumetry is a promising biomarker for early AD detection at a predementia or even presymptomatic stage.
“They suggest that hippocampal subfield volumetry is a promising biomarker for early AD detection at a predementia or even presymptomatic stage”
From Hippocampal subfield volumetry in mild cognitive impairment, Alzheimer's disease and semantic dementia (2013) · cites this paper as Van Leemput et al., 2009
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
Each contradiction between two passages listed here was confirmed by an independent check.
Intensity-correction method cites the tissue-classification reference
MinorThe text attributes its MRI intensity-inhomogeneity correction technique to the 1999b reference, but the reference list identifies the bias-field correction paper as 1999a and the 1999b paper as tissue classification.
“Before segmenting the ROI, we automatically correct the data for MRI intensity inhomogeneities using a previously developed technique (Van Leemput et al., 1999b).”
“20. Van Leemput K Maes F Vandermeulen D Suetens S Automated model-based bias field correction of MR images of the brain IEEE Transac Med Imag 1999a 18 885 896 10628948 21. Van Leemput K Maes F Vandermeulen D Suetens S Automated model-based tissue classification MR images of the brain IEEE Transac Med Imag 1999b 18 897 908 10628949”
Cites
Maximum Likelihood from Incomplete Data Via the <i>EM</i> Algorithm
Dempster, A. P., Laird, N. M., & Rubin, D. B. (1977) · Journal of the Royal Statistical Society Series B: Statistical Methodology · doi:10.1111/j.2517-6161.1977.tb01600.x · Cited by 50,444 · see citing works
Fischl, B., Salat, D. H., Busa, E., Albert, M., Dieterich, M., Haselgrove, C., van der Kouwe, A., Killiany, R., Kennedy, D., Klaveness, S., et al. (2002) · Neuron · doi:10.1016/s0896-6273(02)00569-x · Cited by 8,610 · see citing works
Self-citationAshburner, J., & Friston, K. J. (2005) · NeuroImage · doi:10.1016/j.neuroimage.2005.02.018 · Cited by 7,507 · see citing works
Multimodality image registration by maximization of mutual information
Maes, F., Collignon, A., Vandermeulen, D., Marchal, G., & Suetens, P. (1997) · IEEE Transactions on Medical Imaging · doi:10.1109/42.563664 · Cited by 4,547 · see citing works
Multi-modal volume registration by maximization of mutual information
Wells, W. M., Viola, P. A., Atsumi, H., Nakajima, S., & Kikinis, R. (1996) · Medical Image Analysis · doi:10.1016/s1361-8415(01)80004-9 · Cited by 1,961 · see citing works
Differences in the pattern of hippocampal neuronal loss in normal ageing and Alzheimer's disease
West, M. J., Coleman, P. D., Flood, D. G., & Troncoso, J. C. (1994) · The Lancet · doi:10.1016/s0140-6736(94)92338-8 · Cited by 1,267 · cited 1× · see citing works
“The distinct subregions composing the hippocampus have been implicated in different memory subsystems (Gabrieli et al., 1997; Zeineh et al., 2003), and shown to be differentially affected in normal aging and AD (West et al., 1994; Fukutani et al., 1995; Small et al., 1999; Small et al., 2000).”
BackgroundConstruction of a 3D probabilistic atlas of human cortical structures
Shattuck, D. W., Mirza, M., Adisetiyo, V., Hojatkashani, C., Salamon, G., Narr, K. L., Poldrack, R. A., Bilder, R. M., & Toga, A. W. (2007) · NeuroImage · doi:10.1016/j.neuroimage.2007.09.031 · Cited by 1,208 · see citing works
Automated model-based tissue classification of MR images of the brain
Van Leemput, K., Maes, F., Vandermeulen, D., & Suetens, P. (1999) · IEEE Transactions on Medical Imaging · doi:10.1109/42.811270 · Cited by 1,024 · see citing works
Self-citationMultimodal Image Coregistration and Partitioning—A Unified Framework
Ashburner, J., & Friston, K. J. (1997) · NeuroImage · doi:10.1006/nimg.1997.0290 · Cited by 932 · see citing works
Automated model-based bias field correction of MR images of the brain
Leemput, K. V., Maes, F., Vandermeulen, D., & Suetens, P. (1999) · IEEE Transactions on Medical Imaging · doi:10.1109/42.811268 · Cited by 581 · see citing works
Separate Neural Bases of Two Fundamental Memory Processes in the Human Medial Temporal Lobe
Gabrieli, J. D. E., Brewer, J. B., Desmond, J. E., & Glover, G. H. (1997) · Science · doi:10.1126/science.276.5310.264 · Cited by 569 · see citing works
Dynamics of the Hippocampus During Encoding and Retrieval of Face-Name Pairs
Zeineh, M., Engel, S. A., Thompson, P. M., & Bookheimer, S. Y. (2003) · Science · doi:10.1126/science.1077775 · Cited by 488 · see citing works
32‐channel 3 Tesla receive‐only phased‐array head coil with soccer‐ball element geometry
Wiggins, G. C., Triantafyllou, C., Potthast, A., Reykowski, A., Nittka, M., & Wald, L. L. (2006) · Magnetic Resonance in Medicine · doi:10.1002/mrm.20925 · Cited by 375 · see citing works
Self-citationA Bayesian model for joint segmentation and registration
Pohl, K. M., Fisher, J. S., Grimson, W. E. L., Kikinis, R., & Wells, W. M. (2006) · NeuroImage · doi:10.1016/j.neuroimage.2005.11.044 · Cited by 275 · see citing works
Measurement of hippocampal subfields and age-related changes with high resolution MRI at 4T
Mueller, S. G., Stables, L. A., Du, A., Schuff, N. W., Truran, D., Cashdollar, N. M., & Weiner, M. W. (2006) · Neurobiology of Aging · doi:10.1016/j.neurobiolaging.2006.03.007 · Cited by 266 · see citing works
Burggren, A. C., Zeineh, M., Ekstrom, A. D., Braskie, M. N., Thompson, P. M., Small, G. W., & Bookheimer, S. Y. (2008) · NeuroImage · doi:10.1016/j.neuroimage.2008.03.039 · Cited by 219 · see citing works
Yushkevich, P. A., Avants, B., Pluta, J. B., Das, S. R., Minkoff, D. L., Mechanic-Hamilton, D. J., Glynn, S. M., Pickup, S., LIU, W., & Gee, J. C. (2008) · NeuroImage · doi:10.1016/j.neuroimage.2008.08.042 · Cited by 179 · see citing works
Miller, M. I., Beg, M. F., Ceritoglu, C., & Stark, C. E. L. (2005) · Proceedings of the National Academy of Sciences · doi:10.1073/pnas.0503892102 · Cited by 173 · see citing works
High‐resolution fMRI investigation of the medial temporal lobe
Kirwan, C. B., Jones, C. K., Miller, M. I., & Stark, C. E. L. (2006) · Human Brain Mapping · doi:10.1002/hbm.20331 · Cited by 127 · see citing works
Unfolding the human hippocampus with high resolution structural and functional MRI
Zeineh, M., Engel, S. A., Thompson, P. M., & Bookheimer, S. Y. (2001) · The Anatomical Record · doi:10.1002/ar.1061 · Cited by 107 · see citing works
Application of Cortical Unfolding Techniques to Functional MRI of the Human Hippocampal Region
Zeineh, M., Engel, S. A., & Bookheimer, S. Y. (2000) · NeuroImage · doi:10.1006/nimg.2000.0561 · Cited by 106 · see citing works
Fukutani, Y., Kobayashi, K., Nakamura, I., Watanabe, K., Isaki, K., & Cairns, N. J. (1995) · Neuroscience Letters · doi:10.1016/0304-3940(95)12083-g · Cited by 88 · see citing works
Encoding Probabilistic Brain Atlases Using Bayesian Inference
Leemput, K. V. (2008) · IEEE Transactions on Medical Imaging · doi:10.1109/tmi.2008.2010434 · Cited by 78 · see citing works
Friedman (2000) · doi:10.1002/1097-0029(20001001)51:1
Curio (2000) · doi:10.1002/(sici)1097-0193(200004)9:4
Flagged first, then by citations
Research works printed but not linked: 2
- D'Agostino E, 2006, A Unified Framework for Atlas Based Brain Image Segmentation and Registration. Lecture Notes in Computer Science, 136
- Press WH, 1992, Numerical Recipes in C. The Art of Scientific Computing
Cited by
Sorted by citations
Fischl, B. (2012) · NeuroImage · doi:10.1016/j.neuroimage.2012.01.021 · Cited by 9,526 · see citing works
Self-citationPlasticity in gray and white: neuroimaging changes in brain structure during learning
Zatorre, R. J., Fields, R. D., & Johansen-Berg, H. (2012) · Nature Neuroscience · doi:10.1038/nn.3045 · Cited by 1,685 · see citing works
Teicher, M. H., Anderson, C. M., & Polcari, A. (2012) · Proceedings of the National Academy of Sciences · doi:10.1073/pnas.1115396109 · Cited by 639 · see citing works
Brain development and aging: Overlapping and unique patterns of change
Tamnes, C. K., Walhovd, K. B., Dale, A. M., Østby, Y., Grydeland, H., Richardson, G., Westlye, L. T., Roddey, J. C., Hagler, D. J., Due-Tønnessen, P., et al. (2013) · NeuroImage · doi:10.1016/j.neuroimage.2012.11.039 · Cited by 287 · see citing works
La Joie, R., Perrotin, A., de La Sayette, V., Egret, S., Doeuvre, L., Belliard, S., Eustache, F., Desgranges, B., & Chételat, G. (2013) · NeuroImage: Clinical · doi:10.1016/j.nicl.2013.08.007 · Cited by 290 · see citing works
Pagliaccio, D., Luby, J. L., Bogdan, R., Agrawal, A., Gaffrey, M. S., Belden, A. C., Botteron, K. N., Harms, M. P., & Barch, D. M. (2014) · Neuropsychopharmacology · doi:10.1038/npp.2013.327 · Cited by 205 · see citing works
Subfields of the hippocampal formation at 7 T MRI: In vivo volumetric assessment
Wisse, L. E. M., Gerritsen, L., Zwanenburg, J. J. M., Kuijf, H. J., Luijten, P. R., Biessels, G. J., & Geerlings, M. I. (2012) · NeuroImage · doi:10.1016/j.neuroimage.2012.03.023 · Cited by 183 · see citing works
Booij, L., Szyf, M., Carballedo, A., Frey, E.-M., Morris, D., Dymov, S., Vaisheva, F., Ly, V., Fahey, C., Meaney, J., et al. (2015) · PLOS ONE · doi:10.1371/journal.pone.0119061 · Cited by 174 · see citing works
Imaging the Human Medial Temporal Lobe with High-Resolution fMRI
Carr, V. A., Rissman, J., & Wagner, A. D. (2010) · Neuron · doi:10.1016/j.neuron.2009.12.022 · Cited by 174 · see citing works
Carlesimo, G. A., Piras, F., Orfei, M. D., Iorio, M., Caltagirone, C., & Spalletta, G. (2015) · Alzheimer's & Dementia: Diagnosis, Assessment & Disease Monitoring · doi:10.1016/j.dadm.2014.12.001 · Cited by 159 · see citing works
Thomas, A. G., Dennis, A., Rawlings, N. B., Stagg, C. J., Matthews, L., Morris, M., Kolind, S. H., Foxley, S., Jenkinson, M., Nichols, T. E., et al. (2016) · NeuroImage · doi:10.1016/j.neuroimage.2015.10.090 · Cited by 152 · see citing works
“Segmentation of the hippocampus for each visit was performed on individual subject MPRAGE images (without contrast) in native space using the Freesurfer hippocampal subfield segmentation pipeline (Van Leemput et al., 2009).”
MethodZhao, W., Wang, X., Yin, C., He, M., Li, S., & Han, Y. (2019) · Frontiers in Neuroinformatics · doi:10.3389/fninf.2019.00013 · Cited by 145 · see citing works
Abstract Representations of Location and Facing Direction in the Human Brain
Vass, L. K., & Epstein, R. A. (2013) · The Journal of Neuroscience · doi:10.1523/jneurosci.3873-12.2013 · Cited by 131 · see citing works
Mild Cognitive Impairment: Differential Atrophy in the Hippocampal Subfields
Hanseeuw, B. J., Van Leemput, K., Kavec, M., Grandin, C., Seron, X., & Ivanoiu, A. (2011) · American Journal of Neuroradiology · doi:10.3174/ajnr.a2589 · Cited by 110 · see citing works
Pluta, J., Yushkevich, P., Das, S., & Wolk, D. (2012) · Journal of Alzheimer's Disease · doi:10.3233/jad-2012-111931 · Cited by 107 · see citing works
Chalavi, S., Vissia, E. M., Giesen, M. E., Nijenhuis, E. R. S., Draijer, N., Cole, J. H., Dazzan, P., Pariante, C. M., Madsen, S. K., Rajagopalan, P., et al. (2015) · Human Brain Mapping · doi:10.1002/hbm.22730 · Cited by 100 · see citing works
Automated measurement of hippocampal subfields in PTSD: evidence for smaller dentate gyrus volume
Hayes, J. P., Hayes, S., Miller, D. R., Lafleche, G., Logue, M. W., & Verfaellie, M. (2017) · Journal of Psychiatric Research · doi:10.1016/j.jpsychires.2017.09.007 · Cited by 81 · see citing works
Additive gene-environment effects on hippocampal structure in healthy humans
Rabl, U., Meyer, B. M., Diers, K., Bartova, L., Berger, A., Mandorfer, D., Popovic, A., Scharinger, C., Huemer, J., Kalcher, K., et al. (2014) · Journal of Neuroscience · doi:10.1523/jneurosci.3113-13.2014 · Cited by 68 · see citing works
Mak, E., Gabel, S., Su, L., Williams, G. B., Arnold, R., Passamonti, L., Vazquez Rodríguez, P., Surendranathan, A., Bevan-Jones, W. R., Rowe, J. B., et al. (2017) · International Psychogeriatrics · doi:10.1017/s1041610216002143 · Cited by 66 · see citing works
Hippocampal subfield analysis in medication-naïve female patients with major depressive disorder
Han, K.-M., Won, E., Sim, Y., & Tae, W.-S. (2016) · Journal of Affective Disorders · doi:10.1016/j.jad.2016.01.019 · Cited by 63 · see citing works
7T-guided super-resolution of 3T MRI
Bahrami, K., Shi, F., Rekik, I., Gao, Y., & Shen, D. (2017) · Medical Physics · doi:10.1002/mp.12132 · Cited by 56 · see citing works
Automated Segmentation of Hippocampal Subfields in Drug-Naïve Patients with Alzheimer Disease
Lim, H. K., Hong, S. C., Jung, W. S., Ahn, K. J., Won, W. Y., Hahn, C., Kim, I. S., & Lee, C. U. (2013) · American Journal of Neuroradiology · doi:10.3174/ajnr.a3293 · Cited by 52 · see citing works
Sun, Y., Hu, N., Wang, M., Lu, L., Luo, C., Tang, B., Yao, C., Sweeney, J. A., Gong, Q., Qiu, C., et al. (2023) · Journal of Psychiatry and Neuroscience · doi:10.1503/jpn.220086 · Cited by 48 · see citing works
Advanced Overview of Biomarkers and Techniques for Early Diagnosis of Alzheimer’s Disease
Rani, S., Dhar, S. B., Khajuria, A., Gupta, D., Jaiswal, P. K., Singla, N., Kaur, M., Singh, G., & Barnwal, R. P. (2023) · Cellular and Molecular Neurobiology · doi:10.1007/s10571-023-01330-y · Cited by 48 · see citing works
Li, Y.-D., Dong, H.-B., Xie, G.-M., & Zhang, L.-J. (2013) · American Journal of Alzheimer's Disease & Other Dementias® · doi:10.1177/1533317513494452 · Cited by 42 · see citing works
Altered hippocampal subfield volumes in major depressive disorder with and without anhedonia
Wu, C., Jia, L., Mu, Q., Fang, Z., Hamoudi, H. J. A. S., Huang, M., Hu, S., Zhang, P., Xu, Y., & Lu, S. (2023) · BMC Psychiatry · doi:10.1186/s12888-023-05001-6 · Cited by 29 · see citing works
Yuan, M., Rubin-Falcone, H., Lin, X., Rizk, M. M., Miller, J. M., Sublette, M. E., Oquendo, M. A., Burke, A., Ogden, R. T., & Mann, J. J. (2020) · Journal of Affective Disorders · doi:10.1016/j.jad.2020.03.169 · Cited by 18 · see citing works
Katsuki, A., Watanabe, K., Nguyen, L., Otsuka, Y., Igata, R., Ikenouchi, A., Kakeda, S., Korogi, Y., & Yoshimura, R. (2020) · International Journal of Molecular Sciences · doi:10.3390/ijms21093032 · Cited by 15 · see citing works
- Authors
- Van Leemput, K., Bakkour, A., Benner, T., Wiggins, G., Wald, L. L., Augustinack, J., Dickerson, B. C., Golland, P., & Fischl, B.
- Affiliations
- Harvard University (US)
- Massachusetts Institute of Technology (US)
- ORCID
- Venue
- Hippocampus
- Publisher
- Wiley
- Published
- 2009-04-29
- Volume
- 19
- Issue
- 6
- Pages
- 549-557
- ISSN
- 1050-9631 (ISSN-L)
- 1098-1063
- PMC
- PMC2739884
- OpenAlex
- W1571837886
- Semantic Scholar
- 8281846c96730d51a360b90f3aa3d5dd19b0d408
- Cited by
- 405
- References
- 26
- Citations by year
- 2026: 7 · 2025: 15 · 2024: 11 · 2023: 14 · 2022: 11 · 2021: 18 · 2020: 32 · 2019: 27 · 2018: 37 · 2017: 40 · 2016: 57 · 2015: 40 · 2014: 39 · 2013: 24 · 2012: 18
- FWCI
- 11.76
- Type
- journal article
- Language
- en
- Open access
- bronze
- doi.org
- Locations
- Hippocampus (publishedVersion)
- PubMed (publishedVersion)
- PubMed Central (acceptedVersion)
- DSpace@MIT (Massachusetts Institute of Technology) (submittedVersion, cc-by-nc-sa)
- PubMed Central
- Funders
- Ellison Medical Foundation
- National Institutes of Health (BIRN002, RR14075, EB005149)
- National Institute on Aging (R21AG029840, R01AG029411)
- National Institute of Neurological Disorders and Stroke (NS052585-01, R01NS051826, P41-RR14075, R01NS052585)
- National Institute of Biomedical Imaging and Bioengineering (R01EB001550, R01EB006758, U54EB005149)
- National Center for Research Resources (P41-RR-14075, RR 14075, U24RR021382, R01 RR1659401A1, R01-RR16594, P41RR014075, P41RR013218, BIRN002, R01RR016594, P41 RR13218)
- Topics
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Medical Image Segmentation Techniques
- Life Sciences › Neuroscience › Neurology › Brain Tumor Detection and Classification
- Physical Sciences › Engineering › Biomedical Engineering › Medical Imaging and Analysis
- Keywords
- hippocampal subfields, automatic segmentation, ultra-high resolution MRI, Bayesian inference, hippocampal formation, statistical image models, manual segmentation comparison
- MeSH terms
- Adult, Aged, Aged, 80 and over, Aging, Algorithms, Alzheimer Disease, Automation, Bayes Theorem, Hippocampus, Humans, Magnetic Resonance Imaging, Middle Aged, Organ Size, Young Adult
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-10-10.
Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-10.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
