Limits...
In vivo hippocampal measurement and memory: a comparison of manual tracing and automated segmentation in a large community-based sample.

Cherbuin N, Anstey KJ, Réglade-Meslin C, Sachdev PS - PLoS ONE (2009)

Bottom Line: Few studies have systematically compared the performance of the two techniques.Inspection of 2D and 3D models suggested that this difference was largely due to greater inclusion of boundary voxels by the automated method and variations in subiculum/entorhinal segmentation.This study supports the use of automated measures, based on Freesurfer in this instance, as being sufficiently reliable and valid particularly in the context of larger sample sizes when the research question does not rely on 'true' hippocampal volumes.

View Article: PubMed Central - PubMed

Affiliation: Centre for Mental Health Research, Australian National University, Canberra, Australian Capital Territory, Australia. nicolas.cherbuin@anu.edu.au

ABSTRACT
While manual tracing is the method of choice in measuring hippocampal volume, its time intensive nature and proneness to human error make automated methods attractive, especially when applied to large samples. Few studies have systematically compared the performance of the two techniques. In this study, we measured hippocampal volumes in a large (N = 403) population-based sample of individuals aged 44-48 years using manual tracing by a trained researcher and automated procedure using Freesurfer (http://surfer.nmr.mgh.harvard.edu) imaging suite. Results showed that absolute hippocampal volumes assessed with these methods were significantly different, with automated measures using the Freesurfer software suite being significantly larger, by 23% for the left and 29% for the right hippocampus. The correlation between the two methods varied from 0.61 to 0.80, with lower correlations for hippocampi with visible abnormalities. Inspection of 2D and 3D models suggested that this difference was largely due to greater inclusion of boundary voxels by the automated method and variations in subiculum/entorhinal segmentation. The correlation between left and right hippocampal volumes was very similar by the two methods. The relationship of hippocampal volumes to selected sociodemographic and cognitive variables was not affected by the measurement method, with each measure showing an association with memory performance and suggesting that both were equally valid for this purpose. This study supports the use of automated measures, based on Freesurfer in this instance, as being sufficiently reliable and valid particularly in the context of larger sample sizes when the research question does not rely on 'true' hippocampal volumes.

Show MeSH

Related in: MedlinePlus

Manual tracing (left) and automated segmentation (right) of the hippocampus showing boundary differences (1 & 6: CSF inclusions; 2 greater subicular/entorhinal inclusion; 4 & 5 greater partial volume inclusion).
© Copyright Policy
Related In: Results  -  Collection


getmorefigures.php?uid=PMC2667216&req=5

pone-0005265-g005: Manual tracing (left) and automated segmentation (right) of the hippocampus showing boundary differences (1 & 6: CSF inclusions; 2 greater subicular/entorhinal inclusion; 4 & 5 greater partial volume inclusion).

Mentions: Hippocampal volumes measured using Freesurfer were on average 26% larger than those computed using manual tracings. These findings are consistent with those of Tae et al. [4]. Comparisons of the segmentation protocols used for manual tracing and to create the atlas used in Freesurfer did not identify important differences that could account for the present results. Inspection of 2D and 3D models based on both methodologies suggests that these differences are probably due to two main factors. Firstly, the volumes produced using Freesurfer appear uniformly larger (see Figure 4). A close inspection of single slices (see Figure 5) points to a number of causes including an over-inclusion of boundary voxels composed of both grey and white matter (or cerebrospinal fluid) and misidentification of small pockets of CSF as hippocampal tissue. Secondly, although the manual and automated methodologies both include the subiculum, they differ in their assessment of the subicular/entorhinal and parahippocampal boundary for which there is no absolute cut-off (see Figure 4). As a consequence hippocampal volumes produced in Freesurfer include a substantially greater portion of the subiculum/entorhinal/parahippocampal region than the traced volumes. Since the subiculum may account for as much as 15% of hippocampal volume [2], variations in boundaries may account for a large portion of the difference identified between the manual and automated methods in this study. In itself and given the importance of the subiculum/entorhinal/parahippocampal areas in various cognitive processes including memory, differences in inclusion of these structures is not necessarily a negative factor, provided it is done consistently, but it needs to be taken into consideration in the interpretation of results based on automated segmentation with Freesurfer.


In vivo hippocampal measurement and memory: a comparison of manual tracing and automated segmentation in a large community-based sample.

Cherbuin N, Anstey KJ, Réglade-Meslin C, Sachdev PS - PLoS ONE (2009)

Manual tracing (left) and automated segmentation (right) of the hippocampus showing boundary differences (1 & 6: CSF inclusions; 2 greater subicular/entorhinal inclusion; 4 & 5 greater partial volume inclusion).
© Copyright Policy
Related In: Results  -  Collection

Show All Figures
getmorefigures.php?uid=PMC2667216&req=5

pone-0005265-g005: Manual tracing (left) and automated segmentation (right) of the hippocampus showing boundary differences (1 & 6: CSF inclusions; 2 greater subicular/entorhinal inclusion; 4 & 5 greater partial volume inclusion).
Mentions: Hippocampal volumes measured using Freesurfer were on average 26% larger than those computed using manual tracings. These findings are consistent with those of Tae et al. [4]. Comparisons of the segmentation protocols used for manual tracing and to create the atlas used in Freesurfer did not identify important differences that could account for the present results. Inspection of 2D and 3D models based on both methodologies suggests that these differences are probably due to two main factors. Firstly, the volumes produced using Freesurfer appear uniformly larger (see Figure 4). A close inspection of single slices (see Figure 5) points to a number of causes including an over-inclusion of boundary voxels composed of both grey and white matter (or cerebrospinal fluid) and misidentification of small pockets of CSF as hippocampal tissue. Secondly, although the manual and automated methodologies both include the subiculum, they differ in their assessment of the subicular/entorhinal and parahippocampal boundary for which there is no absolute cut-off (see Figure 4). As a consequence hippocampal volumes produced in Freesurfer include a substantially greater portion of the subiculum/entorhinal/parahippocampal region than the traced volumes. Since the subiculum may account for as much as 15% of hippocampal volume [2], variations in boundaries may account for a large portion of the difference identified between the manual and automated methods in this study. In itself and given the importance of the subiculum/entorhinal/parahippocampal areas in various cognitive processes including memory, differences in inclusion of these structures is not necessarily a negative factor, provided it is done consistently, but it needs to be taken into consideration in the interpretation of results based on automated segmentation with Freesurfer.

Bottom Line: Few studies have systematically compared the performance of the two techniques.Inspection of 2D and 3D models suggested that this difference was largely due to greater inclusion of boundary voxels by the automated method and variations in subiculum/entorhinal segmentation.This study supports the use of automated measures, based on Freesurfer in this instance, as being sufficiently reliable and valid particularly in the context of larger sample sizes when the research question does not rely on 'true' hippocampal volumes.

View Article: PubMed Central - PubMed

Affiliation: Centre for Mental Health Research, Australian National University, Canberra, Australian Capital Territory, Australia. nicolas.cherbuin@anu.edu.au

ABSTRACT
While manual tracing is the method of choice in measuring hippocampal volume, its time intensive nature and proneness to human error make automated methods attractive, especially when applied to large samples. Few studies have systematically compared the performance of the two techniques. In this study, we measured hippocampal volumes in a large (N = 403) population-based sample of individuals aged 44-48 years using manual tracing by a trained researcher and automated procedure using Freesurfer (http://surfer.nmr.mgh.harvard.edu) imaging suite. Results showed that absolute hippocampal volumes assessed with these methods were significantly different, with automated measures using the Freesurfer software suite being significantly larger, by 23% for the left and 29% for the right hippocampus. The correlation between the two methods varied from 0.61 to 0.80, with lower correlations for hippocampi with visible abnormalities. Inspection of 2D and 3D models suggested that this difference was largely due to greater inclusion of boundary voxels by the automated method and variations in subiculum/entorhinal segmentation. The correlation between left and right hippocampal volumes was very similar by the two methods. The relationship of hippocampal volumes to selected sociodemographic and cognitive variables was not affected by the measurement method, with each measure showing an association with memory performance and suggesting that both were equally valid for this purpose. This study supports the use of automated measures, based on Freesurfer in this instance, as being sufficiently reliable and valid particularly in the context of larger sample sizes when the research question does not rely on 'true' hippocampal volumes.

Show MeSH
Related in: MedlinePlus