CoChat
Sign Up
Cited by 110Advanced Neuroimaging Techniques and ApplicationsFunctional Brain Connectivity StudiesAdvanced MRI Techniques and Applications

The DOI matches its Crossref record; 57 of the 57 references it cites were found in a registry or an index.

Fusion in diffusion MRI for improved fibre orientation estimation: An application to the 3T and 7T data of the Human Connectome Project

Open access
Sotiropoulos, Stamatios N., Hernández-Fernández, Moisés, Vu, An T., Andersson, Jesper L., Moeller, Steen, Yacoub, Essa, Lenglet, Christophe, Ugurbil, Kamil, Behrens, Timothy E.J., Jbabdi, Saad·2016·NeuroImage
DOI 10.1016/j.neuroimage.2016.04.014PMC PMC6318224
Open this paper in CoChat — free accountOpen PDF

Abstract

Determining the acquisition parameters in diffusion magnetic resonance imaging (dMRI) is governed by a series of trade-offs. Images of lower resolution have less spatial specificity but higher signal to noise ratio (SNR). At the same time higher angular contrast, important for resolving complex fibre patterns, also yields lower SNR. Considering these trade-offs, the Human Connectome Project (HCP) acquires high quality dMRI data for the same subjects at different field strengths (3T and 7T), which are publically released. Due to differences in the signal behavior and in the underlying scanner hardware, the HCP 3T and 7T data have complementary features in k- and q-space. The 3T dMRI has higher angular contrast and resolution, while the 7T dMRI has higher spatial resolution. Given the availability of these datasets, we explore the idea of fusing them together with the aim of combining their benefits. We extend a previously proposed data-fusion framework and apply it to integrate both datasets from the same subject into a single joint analysis. We use a generative model for performing parametric spherical deconvolution and estimate fibre orientations by simultaneously using data acquired under different protocols. We illustrate unique features from each dataset and how they are retained after fusion. We further show that this allows us to complement benefits and improve brain connectivity analysis compared to analyzing each of the datasets individually.

Overview

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.1016/j.neuroimage.2016.04.014

Cites

57 of 57 references verified0 citation errors · 0 not checkable

Statistics

Figures

Fig. 1 A) Mean and standard deviation of the anisotropy ratio (λR = λradial/λaxial) across ten different HCP subjects. The values for each subject were obtained using the DTI eigenvalues in a region depicting the most anisotropic voxels in the corpus callosum (red mask in B). The green dashed lines correspond to the average values used for the hyperparameters of the λR prior distribution (μλ = 0.13, σλ = 0.03). B) Axial and sagittal views of the mask depicting the most anisotropic voxels of the corpus callosum. The midbody of the corpus callosum was identified in the Johns Hopkins University white matter atlas (Wakana et al., 2007), available in FSL. The mask was eroded once and voxels with the top 50% FA values within this mask were retained. The mask is shown in red superimposed on an FA map.
Fig. 1 A) Mean and standard deviation of the anisotropy ratio (λR = λradial/λaxial) across ten different HCP subjects. The values for each subject were obtained using the DTI eigenvalues in a region depicting the most anisotropic voxels in the corpus callosum (red mask in B). The green dashed lines correspond to the average values used for the hyperparameters of the λR prior distribution (μλ = 0.13, σλ = 0.03). B) Axial and sagittal views of the mask depicting the most anisotropic voxels of the corpus callosum. The midbody of the corpus callosum was identified in the Johns Hopkins University white matter atlas (Wakana et al., 2007), available in FSL. The mask was eroded once and voxels with the top 50% FA values within this mask were retained. The mask is shown in red superimposed on an FA map.CC BY · PubMed Central
Fig. 2 DTI maps for 3T and 7T data of the same subject (HCP Subject ID: 158035). Top row: Fractional anisotropy (FA) maps (axial, coronal and sagittal views). Notice the effect of B1 inhomogeneities for the 7T that lead to poor SNR and noisy FA estimates at the inferior temporal regions (evident in the coronal views). These artifacts are present for the 7T data, but efforts have been taken to minimise them (Vu et al., 2015). Bottom row: DTI principal fibre orientations (coronal zoomed view of the area delineated by the yellow box). The orientations are RGB color-coded (Red: Left–Right, Green: Anterior–Posterior, Blue: Superior–Inferior) and superimposed on the structural T1w image. The pial surface and the WM/GM boundary surface are also shown.
Fig. 2 DTI maps for 3T and 7T data of the same subject (HCP Subject ID: 158035). Top row: Fractional anisotropy (FA) maps (axial, coronal and sagittal views). Notice the effect of B1 inhomogeneities for the 7T that lead to poor SNR and noisy FA estimates at the inferior temporal regions (evident in the coronal views). These artifacts are present for the 7T data, but efforts have been taken to minimise them (Vu et al., 2015). Bottom row: DTI principal fibre orientations (coronal zoomed view of the area delineated by the yellow box). The orientations are RGB color-coded (Red: Left–Right, Green: Anterior–Posterior, Blue: Superior–Inferior) and superimposed on the structural T1w image. The pial surface and the WM/GM boundary surface are also shown.CC BY · PubMed Central
Fig. 3 RubiX model predictions of both LR (3T) and HR (7T) datasets for three different locations (in the midbody of the corpus callosum, in the centrum semiovale and in cortical grey matter). For each model parameter, the mode of the estimated posterior distribution was used to obtain model predictions. The volumes have been grouped into b-shells. Within each shell, the volumes have been rearranged according to the dot product between the corresponding gradient direction and the principal DTI eigenvector.
Fig. 3 RubiX model predictions of both LR (3T) and HR (7T) datasets for three different locations (in the midbody of the corpus callosum, in the centrum semiovale and in cortical grey matter). For each model parameter, the mode of the estimated posterior distribution was used to obtain model predictions. The volumes have been grouped into b-shells. Within each shell, the volumes have been rearranged according to the dot product between the corresponding gradient direction and the principal DTI eigenvector.CC BY · PubMed Central
Fig. 4 RubiX model predictions of different measurement volumes in the HR (7T) and LR (3T) data. The top row shows the actual measurements (the 3T data have been downsampled, as used in the estimation). The middle row shows the predictions. To obtain model predictions, the mode of the estimated posterior distribution was used for each model parameter. The bottom row shows the difference of the prediction minus the measurement expressed as a percentage of the measurement. The pial surface is shown as a grey outline (HCP Subject ID: 102311).
Fig. 4 RubiX model predictions of different measurement volumes in the HR (7T) and LR (3T) data. The top row shows the actual measurements (the 3T data have been downsampled, as used in the estimation). The middle row shows the predictions. To obtain model predictions, the mode of the estimated posterior distribution was used for each model parameter. The bottom row shows the difference of the prediction minus the measurement expressed as a percentage of the measurement. The pial surface is shown as a grey outline (HCP Subject ID: 102311).CC BY · PubMed Central
Fig. 5 Fibre orientations in the centrum semiovale (coronal views) estimated using (from left to right): Voxel-wise deconvolution on the 3T alone (estimates at nominal spatial resolution (1.25 mm)3, Voxel-wise deconvolution on the 7T alone (estimates at nominal spatial resolution (1.05 mm)3), RubiX deconvolution on both 3T and 7T (estimates at nominal spatial resolution (1.05 mm)3). The two rows correspond to two different subjects. The black contour represents the WM/GM boundary surface. The vectors have been RGB color-coded and their norm is modulated by the respective volume fractions. Only fibres with volume fraction f > 5% are shown.
Fig. 5 Fibre orientations in the centrum semiovale (coronal views) estimated using (from left to right): Voxel-wise deconvolution on the 3T alone (estimates at nominal spatial resolution (1.25 mm)3, Voxel-wise deconvolution on the 7T alone (estimates at nominal spatial resolution (1.05 mm)3), RubiX deconvolution on both 3T and 7T (estimates at nominal spatial resolution (1.05 mm)3). The two rows correspond to two different subjects. The black contour represents the WM/GM boundary surface. The vectors have been RGB color-coded and their norm is modulated by the respective volume fractions. Only fibres with volume fraction f > 5% are shown.CC BY · PubMed Central
Fig. 6 Number of crossing fibres (top row) and uncertainty of estimated orientations (bottom row) obtained using four different methods: voxel-wise deconvolution at 3T (red), voxel-wise deconvolution at 7T (blue), RubiX deconvolution using both 3T and 7T and w = 1 for model selection (Rbx1, purple) and RubiX deconvolution using w = 0.8 (Rbx, black). Notice the difference in the spatial resolution, on which each method operates and estimates fibre orientations. Two ROIs have been considered, one in the corona radiata (obtained from the Johns Hopkins University white matter atlas (Wakana et al., 2007)) and one covering the white matter. The bars correspond to the mean and standard deviation across 10 subjects. For each subject and each ROI, the mean values were obtained. Only fibre orientations with f > 5% were considered. The percentages at the top row are expressed as fractions of the total ROI volume. The uncertainties at the bottom correspond to the 95% cone of uncertainty (Jones, 2003).
Fig. 6 Number of crossing fibres (top row) and uncertainty of estimated orientations (bottom row) obtained using four different methods: voxel-wise deconvolution at 3T (red), voxel-wise deconvolution at 7T (blue), RubiX deconvolution using both 3T and 7T and w = 1 for model selection (Rbx1, purple) and RubiX deconvolution using w = 0.8 (Rbx, black). Notice the difference in the spatial resolution, on which each method operates and estimates fibre orientations. Two ROIs have been considered, one in the corona radiata (obtained from the Johns Hopkins University white matter atlas (Wakana et al., 2007)) and one covering the white matter. The bars correspond to the mean and standard deviation across 10 subjects. For each subject and each ROI, the mean values were obtained. Only fibre orientations with f > 5% were considered. The percentages at the top row are expressed as fractions of the total ROI volume. The uncertainties at the bottom correspond to the 95% cone of uncertainty (Jones, 2003).CC BY · PubMed Central
Fig. 7 Angle between surface tangent and fibre orientations at the inflated WM/GM boundary surface. Orientations have been estimated as before using voxel-wise deconvolution on the 3T alone, voxel-wise deconvolution on the 7T alone, RubiX deconvolution on both 3T and 7T. For every surface vertex, the maximum dot product between fibre orientations (with volume fraction f > 5%) at this location (within a voxel size distance from the vertex location and on either side of the surface) and the surface normal is computed. This is then converted to the colour-code angle shown on the inflated surface. To aid visualisation of these qualitative illustrations, Gaussian smoothing on the surface with 1 mm FWHM was performed for all cases. (HCP Subject ID: 158035).
Fig. 7 Angle between surface tangent and fibre orientations at the inflated WM/GM boundary surface. Orientations have been estimated as before using voxel-wise deconvolution on the 3T alone, voxel-wise deconvolution on the 7T alone, RubiX deconvolution on both 3T and 7T. For every surface vertex, the maximum dot product between fibre orientations (with volume fraction f > 5%) at this location (within a voxel size distance from the vertex location and on either side of the surface) and the surface normal is computed. This is then converted to the colour-code angle shown on the inflated surface. To aid visualisation of these qualitative illustrations, Gaussian smoothing on the surface with 1 mm FWHM was performed for all cases. (HCP Subject ID: 158035).CC BY · PubMed Central
Fig. 8 Normalised histograms of angles between fibre orientations and surface tangent at three subregions of the WM/GM boundary surface: gyral crows (top), gyral walls (middle) and sulcal fundi (bottom). Orientations have been estimated as before using voxel-wise deconvolution on the 3T alone (red), voxel-wise deconvolution on the 7T alone (blue), RubiX deconvolution on both 3T and 7T (black). For every surface vertex, the maximum dot product between fibreorientations (with volume fraction f > 5%) at this location and the surface normal is computed. This is then converted to the angles shown. For every bin of the histograms, the mean angle across ten subjects is shown. The error bars indicate the standard deviation across these subjects.
Fig. 8 Normalised histograms of angles between fibre orientations and surface tangent at three subregions of the WM/GM boundary surface: gyral crows (top), gyral walls (middle) and sulcal fundi (bottom). Orientations have been estimated as before using voxel-wise deconvolution on the 3T alone (red), voxel-wise deconvolution on the 7T alone (blue), RubiX deconvolution on both 3T and 7T (black). For every surface vertex, the maximum dot product between fibreorientations (with volume fraction f > 5%) at this location and the surface normal is computed. This is then converted to the angles shown. For every bin of the histograms, the mean angle across ten subjects is shown. The error bars indicate the standard deviation across these subjects.CC BY · PubMed Central
Fig. 9 A) Correlations of mean path probabilities on the WM/GM boundary surface with gyral height (“sulc” map obtained from FreeSurfer). Violin plots illustrate the distribution of these correlations across ten subjects and for three different methods (3T, 7T, Rbx). The dashed lines within the plots show the median of the distribution and the dotted lines the inter-quartile range. The individual points are also shown and the solid lines connect the points corresponding to the same subject, across the three different methods. The mean path probabilities were estimated using tractography. As shown on the left column, probabilistic tractography was seeded at every brain tissue voxel C (~ 190,000 seeds, 3000 samples per seed, curvature threshold = 80°, Pial surface termination mask, WM/GM boundary surface allowed to be crossed only twice). Path probabilities were then recorded between pairs of 60,000 vertices on the WM/GM boundary surface to give a comprehensive connectome matrix. The mean path probability of every vertex was obtained by taking the average across the rows of the connectome. The logarithm of the path probabilities was correlated with the gyral height at the respective vertex locations. Gyral height is high for gyral crowns and low for sulcal fundi. B) Mean path probabilities at different regions of the WM/GM boundary surface (sulcal fundi, gyral walls), normalised to the corresponding mean path probability at the gyral crowns. The mean path probability values obtained on the WM/GM surface as in (A) were grouped into the three subregions described in Fig. 8. The average across these subregions was obtained and the ratio of these values at the sulcal fundi and the gyral walls to the value at the gyral crowns is presented. The bars show the mean and standard deviation of the ratios across ten subjects.
Fig. 9 A) Correlations of mean path probabilities on the WM/GM boundary surface with gyral height (“sulc” map obtained from FreeSurfer). Violin plots illustrate the distribution of these correlations across ten subjects and for three different methods (3T, 7T, Rbx). The dashed lines within the plots show the median of the distribution and the dotted lines the inter-quartile range. The individual points are also shown and the solid lines connect the points corresponding to the same subject, across the three different methods. The mean path probabilities were estimated using tractography. As shown on the left column, probabilistic tractography was seeded at every brain tissue voxel C (~ 190,000 seeds, 3000 samples per seed, curvature threshold = 80°, Pial surface termination mask, WM/GM boundary surface allowed to be crossed only twice). Path probabilities were then recorded between pairs of 60,000 vertices on the WM/GM boundary surface to give a comprehensive connectome matrix. The mean path probability of every vertex was obtained by taking the average across the rows of the connectome. The logarithm of the path probabilities was correlated with the gyral height at the respective vertex locations. Gyral height is high for gyral crowns and low for sulcal fundi. B) Mean path probabilities at different regions of the WM/GM boundary surface (sulcal fundi, gyral walls), normalised to the corresponding mean path probability at the gyral crowns. The mean path probability values obtained on the WM/GM surface as in (A) were grouped into the three subregions described in Fig. 8. The average across these subregions was obtained and the ratio of these values at the sulcal fundi and the gyral walls to the value at the gyral crowns is presented. The bars show the mean and standard deviation of the ratios across ten subjects.CC BY · PubMed Central
Fig. 10 Somatotopy of motor projections in the internal capsule and the thalamus. Tractography was seeded from the foot, hand and face area of the primary motor cortex, as shown on the left. Axial views of the path distributions are shown for the different methods at two different Z positions. All results are shown in the standard MNI 1 mm space to allow direct comparisons. Path distributions for the foot and face area have been thresholded at 1% (3% threshold has been used for the hand area).
Fig. 10 Somatotopy of motor projections in the internal capsule and the thalamus. Tractography was seeded from the foot, hand and face area of the primary motor cortex, as shown on the left. Axial views of the path distributions are shown for the different methods at two different Z positions. All results are shown in the standard MNI 1 mm space to allow direct comparisons. Path distributions for the foot and face area have been thresholded at 1% (3% threshold has been used for the hand area).CC BY · PubMed Central
Fig. 11 A) Maximum intensity projections along the coronal plane of the distribution of the limbic paths connecting subgenual white matter to the amygdala. Each column corresponds to probabilistic tractography results using fibre orientations estimated from the three different methods (voxelwise 3T deconvolution, voxelwise 7T deconvolution and RubiX deconvolution). The two rows correspond to different subjects (HCP Subject ID: 158035 — top, 109123 — bottom). B) Mean b = 0 image of the 7T data for the two subjects presented in (A) (HCP Subject ID: 158035 — top, 109123 — bottom). The red outlines represent the pial surface. Subjects with larger brain sizes have larger B1 inhomogeneity artifacts. These lead to more severe signal dropouts in the temporal lobes (yellow arrows).
Fig. 11 A) Maximum intensity projections along the coronal plane of the distribution of the limbic paths connecting subgenual white matter to the amygdala. Each column corresponds to probabilistic tractography results using fibre orientations estimated from the three different methods (voxelwise 3T deconvolution, voxelwise 7T deconvolution and RubiX deconvolution). The two rows correspond to different subjects (HCP Subject ID: 158035 — top, 109123 — bottom). B) Mean b = 0 image of the 7T data for the two subjects presented in (A) (HCP Subject ID: 158035 — top, 109123 — bottom). The red outlines represent the pial surface. Subjects with larger brain sizes have larger B1 inhomogeneity artifacts. These lead to more severe signal dropouts in the temporal lobes (yellow arrows).CC BY · PubMed Central
Fig. 12 Replication of the organisation pattern of prefrontal cortex commissural projections using HCP data. A) The ventral prefrontal cortex (vPFC) has been divided into 13 subregions according to Ongur et al. (2003). The regions are shown from a sagittal perspective, as well as from below (A: anterior, P: posterior, S: superior, I: inferior, M: medial, L: lateral). The organisation pattern predicts that the more lateral the position of a subregion is, the more superior in the corpus callosum (CC) its commissural projection will be. Tracers injected in the vPFC, as well as macaque post-mortem tractography follow this pattern with a very strong correlation between the X position of the centre of gravity (COG) in the vPFC and the Z position of the COG of the callosal projection (Jbabdi et al., 2013). B) The coefficient of determination R2 between the vPFC X and the midsaggital CC Z position, when performing tractography using the fibre orientations estimated from the three different approaches. Mean and standard deviation shown across ten subjects. For each subject, the COG is computed for each vPFC mask (since these are binary masks, the COG is simply the mean of the coordinates of all voxels contained in the mask). Also, for each mask the respective callosal path distribution is obtained using tractography. The COG of the distribution is obtained on the midsagittal plane (since this is a path distribution, the COG is the centre of mass of this distribution, i.e. coordinates are weighted by the respective probabilities). The squared correlation coefficient between the X coordinate of vPFC COG and the Z coordinate of the respective CC COG is then computed. The barplots show these values across ten subjects. RubiX results have on average the highest and less variable R2 compared to the other two methods. C) Mean coordinates across ten subjects of the COG of the midsaggital CC projections from each vPFC subregion when using RubiX. D) Mean and standard deviation across subjects of coordinates of the midsaggital CC projections' COG when using RubiX. The grey squares delineate the anterior part of the corpus callosum. Coordinates are all reported in MNI space. For all tractography a waypoint was used for the anterior CC. For each vPFC seed, the rest subregions were used as exclusion masks.
Fig. 12 Replication of the organisation pattern of prefrontal cortex commissural projections using HCP data. A) The ventral prefrontal cortex (vPFC) has been divided into 13 subregions according to Ongur et al. (2003). The regions are shown from a sagittal perspective, as well as from below (A: anterior, P: posterior, S: superior, I: inferior, M: medial, L: lateral). The organisation pattern predicts that the more lateral the position of a subregion is, the more superior in the corpus callosum (CC) its commissural projection will be. Tracers injected in the vPFC, as well as macaque post-mortem tractography follow this pattern with a very strong correlation between the X position of the centre of gravity (COG) in the vPFC and the Z position of the COG of the callosal projection (Jbabdi et al., 2013). B) The coefficient of determination R2 between the vPFC X and the midsaggital CC Z position, when performing tractography using the fibre orientations estimated from the three different approaches. Mean and standard deviation shown across ten subjects. For each subject, the COG is computed for each vPFC mask (since these are binary masks, the COG is simply the mean of the coordinates of all voxels contained in the mask). Also, for each mask the respective callosal path distribution is obtained using tractography. The COG of the distribution is obtained on the midsagittal plane (since this is a path distribution, the COG is the centre of mass of this distribution, i.e. coordinates are weighted by the respective probabilities). The squared correlation coefficient between the X coordinate of vPFC COG and the Z coordinate of the respective CC COG is then computed. The barplots show these values across ten subjects. RubiX results have on average the highest and less variable R2 compared to the other two methods. C) Mean coordinates across ten subjects of the COG of the midsaggital CC projections from each vPFC subregion when using RubiX. D) Mean and standard deviation across subjects of coordinates of the midsaggital CC projections' COG when using RubiX. The grey squares delineate the anterior part of the corpus callosum. Coordinates are all reported in MNI space. For all tractography a waypoint was used for the anterior CC. For each vPFC seed, the rest subregions were used as exclusion masks.CC BY · PubMed Central

Verification

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

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?

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Citations

Cites

Show all 57 references

Cited by

Sorted by citations

Related

See all

Metadata

Authors
Sotiropoulos, S. N., Hernández-Fernández, M., Vu, A. T., Andersson, J. L., Moeller, S., Yacoub, E., Lenglet, C., Ugurbil, K., Behrens, T. E. J., & Jbabdi, S.
Affiliations
  • University of Oxford (GB)
  • University of Minnesota (US)
  • Wellcome Centre for Human Neuroimaging (GB)
Corresponding author
Sotiropoulos, Stamatios N.
ORCID
Venue
NeuroImage
Publisher
Elsevier BV
Published
2016-07
Volume
134
Pages
396-409
ISSN
1053-8119 (ISSN-L)
DOI
10.1016/j.neuroimage.2016.04.014
PMC
PMC6318224
OpenAlex
W2328941635
Cited by
110
References
57
Citations by year
2026: 2 · 2025: 14 · 2024: 15 · 2023: 6 · 2022: 13 · 2021: 13 · 2020: 13 · 2019: 8 · 2018: 11 · 2017: 10 · 2016: 4
FWCI
9.55
Type
journal article
Language
en
Open access
Licence
Locations
  • NeuroImage (publishedVersion, cc-by)
  • Europe PMC (PubMed Central) (publishedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • Repository@Nottingham (University of Nottingham) (submittedVersion, cc-by)
Funders
  • Human Connectome Project (1U54MH091657-01)
  • NIH (P41 EB015894)
  • UK EPSRC (EP/L023067/1)
  • UK MRC (MR/L009013/1)
Topics
  • Health Sciences › Medicine › Radiology, Nuclear Medicine and Imaging › Advanced Neuroimaging Techniques and Applications
  • Life Sciences › Neuroscience › Cognitive Neuroscience › Functional Brain Connectivity Studies
  • Health Sciences › Medicine › Radiology, Nuclear Medicine and Imaging › Advanced MRI Techniques and Applications
Keywords
diffusion MRI, data fusion, Human Connectome Project, brain connectivity, generative models, high-field MRI
MeSH terms
Adult, Algorithms, Brain, Female, Humans, Image Enhancement, Image Interpretation, Computer-Assisted, Male, Sensitivity and Specificity, Subtraction Technique, Reproducibility of Results, Diffusion Tensor Imaging, Connectome, White Matter

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.

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