ArticlePDF Available

Abstract and Figures

Tractography based on non-invasive diffusion imaging is central to the study of human brain connectivity. To date, the approach has not been systematically validated in ground truth studies. Based on a simulated human brain data set with ground truth tracts, we organized an open international tractography challenge, which resulted in 96 distinct submissions from 20 research groups. Here, we report the encouraging finding that most state-of-the-art algorithms produce tractograms containing 90% of the ground truth bundles (to at least some extent). However, the same tractograms contain many more invalid than valid bundles, and half of these invalid bundles occur systematically across research groups. Taken together, our results demonstrate and confirm fundamental ambiguities inherent in tract reconstruction based on orientation information alone, which need to be considered when interpreting tractography and connectivity results. Our approach provides a novel framework for estimating reliability of tractography and encourages innovation to address its current limitations.
No caption available
This content is subject to copyright. Terms and conditions apply.
The challenge of mapping the human connectome
based on diffusion tractography
Klaus H. Maier-Hein
Tractography based on non-invasive diffusion imaging is central to the study of human brain
connectivity. To date, the approach has not been systematically validated in ground truth
studies. Based on a simulated human brain data set with ground truth tracts, we organized an
open international tractography challenge, which resulted in 96 distinct submissions from 20
research groups. Here, we report the encouraging nding that most state-of-the-art algo-
rithms produce tractograms containing 90% of the ground truth bundles (to at least some
extent). However, the same tractograms contain many more invalid than valid bundles, and
half of these invalid bundles occur systematically across research groups. Taken together, our
results demonstrate and conrm fundamental ambiguities inherent in tract reconstruction
based on orientation information alone, which need to be considered when interpreting
tractography and connectivity results. Our approach provides a novel framework for esti-
mating reliability of tractography and encourages innovation to address its current
Corrected: Author correction
DOI: 10.1038/s41467-017-01285-x OPEN
Correspondence and requests for materials should be addressed to K.H.M.-H. (email:
or to M.D. (email:
#A full list of authors and their afiations appears at the end of the paper
NATURE COMMUNICATIONS |8: 1349 |DOI: 10.1038/s41467-017-01285-x | 1
Content courtesy of Springer Nature, terms of use apply. Rights reserved
Tractography, a computational reconstruction method based
on diffusion-weighted magnetic resonance imaging (DWI),
attempts to reveal the trajectories of white matter pathways
in vivo and to infer the underlying structural connectome of the
human brain1. Numerous algorithms for tractography have been
developed and applied to connectome research in the eld of
neuroscience2and psychiatry3. Given the broad interest in this
approach, advantages and shortcomings of tractography have
been addressed using a wide range of approaches1,48. Particu-
larly, in vivo tractography of the human brain has been evaluated
by subjective assessment of plausibility9,10 or qualitative visual
agreement with post-mortem Klingler-like dissections11,12.
Reproducibility13 or data prediction errors1416 have been eval-
uated in the context of tractography model verication. However,
these evaluations cannot validate the accuracy of reconstructions
due to the lack of ground truth information17. Ex vivo imaging
and tracing1723 or physically2430 and numerically simulated
phantoms3134 allow validation to some extent, and in specic
circumstances such as basic ber congurations. The nervous
system, however, is complex and precise ground truth informa-
tion on the trajectories of pathways and their origins, as well as
terminations in the human brain is lacking. This makes it hard to
obtain quantitative and comprehensive reliability estimations of
tractography and to determine which discoveries are reliable
when regarding brain connectivity in health and disease.
State-of-the-art tractography algorithms are driven by local
orientation elds estimated from DWI, representing the local
tangent direction to the white matter tract of interest1.
Conceptually, the principle of inferring connectivity from local
orientation elds can lead to problems as soon as pathways
overlap, cross, branch, and have complex geometries7,35,36. Since
the invention of diffusion tractography, these problems have been
discussed in schematic representations or theoretical
arguments7,8,37, but have not yet been quantied in brain
imaging. To determine the current state of the art in tractography,
we organized an international tractography competition
( We employed
simulated DWI of a brain-like geometry as a novel reliability
estimation method that allowed for a quantitative evaluation
of the submissions based on the Tractometer connectivity
At the closing of the competition, we evaluated 96 distinct
tractography pipelines submitted by 20 different research groups,
in order to assess how well the algorithms were able to reproduce
the known connectivity. We also assessed essential processing
steps to pinpoint critical aws that many current pipelines have in
common. An important positive nding is that most proposed
algorithms are able to produce tractograms containing 90% of the
ground truth bundles, recovering about one-third of their volu-
metric extent. At the same time, most algorithms produce large
amounts of false-positive bundles, even though they are not part
of the ground truth. Results do not improve when employing
higher-quality data or even using the gold standard eld of local
tract orientations at high spatial resolution. The ndings highlight
that novel technological and conceptual developments are needed
to address these limitations.
Real MRI image Bundle
bundle model
synthetic model
Synthetic diffusion-weighted
and T1 image
Fig. 1 Overview of synthetic data set. The top row summarizes the phantom generation process. The simulated images are generated from 25 major
bundles, which are shown in the bottom part of the gure. These were manually segmented from a whole-brain tractogram of a HCP subject and include
the CC, cingulum (Cg), fornix (Fx), anterior commissure (CA), optic radiation (OR), posterior commissure (CP), inferior cerebellar peduncle (ICP), middle
cerebellar peduncle (MCP), superior cerebellar peduncle (SCP), parieto-occipital pontine tract (POPT), cortico-spinal tract (CST), frontopontine tracts
(FPT), ILF, UF, and SLF. The connectivity plot in the middle shows the phantom design. The segment positions correspond to the involved endpoint region
(from top to bottom: frontal lobe, temporal lobe, parietal lobe, occipital lobe, subcortical region, cerebellum, brain stem). The radial segment length and the
connection number in the plot are chosen according to the volume of the respective bundle endpoint region. Abbreviations: right (R) and left (L)
hemisphere, head (H) and tail (T) of each respective bundle
ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/s41467-017-01285-x
2NATURE COMMUNICATIONS |8: 1349 |DOI: 10.1038/s41467-017-01285-x |
Content courtesy of Springer Nature, terms of use apply. Rights reserved
Data sets and submissions. Prior investigations of tractography
methodology have chosen articial ber geometries to construct
synthetic ground truth models26,38. Here, we dened our models
based on the ber bundle geometry of a high-quality Human
Connectome Project (HCP) data set that was constructed from
multiple whole-brain global tractography maps39 (Fig. 1). Following
the concepts introduced in ref. 40,anexpertradiologist
(B.S.) extracted 25 major tracts (i.e., bundles of streamlines) from
the tractogram. This ground truth data set included association,
projection, and commissural tracts that have been previously
described using post-mortem anatomical and electrophysiological
methods41. In total the tracts occupy 71% of the white matter
volume in the human brain. The data set features a brain-like
macro-structure of long-range connections, mimicking in vivo DWI
clinical-like acquisitions based on a simulated diffusion signal. An
additional anatomical image with T1-like contrast was simulated as
a reference. The nal data sets and all les necessary to perform the
simulation are openly available (see Data availability).
Twenty research groups with extensive expertise in diffusion
imaging from 12 countries (Fig. 2a) participated in the competi-
tion and submitted a total of 96 tractograms (see Data availability)
generated using a large variety of tractography pipelines with
different pre-processing, local reconstruction, tractography, and
post-processing algorithms (Fig. 2b, Supplementary Note 1).
Performance metrics and evaluation. The Tractometer con-
nectivity metrics38 were used for a quantitative evaluation of the
submissions. Based on the known ground truth bundles, we
calculated true positives, corresponding to the valid connection
(VC) ratio, that is, the proportion of streamlines connecting valid
end points and the associated number of valid bundles (VB),
where a bundle is a group of streamlines. We also computed false
positives, corresponding to the invalid connection (IC) ratio and
the associated number of invalid bundles (IB), as well as recon-
structed volumes, based on the bundle volumetric overlap (OL)
and volumetric overreach (OR) in percent (see Methodssection
for details and Supplementary Figs. 1,2for alternative metrics).
Tractograms contained most of the ground truth bundles. The
volumetric reconstruction of the existing bundles varied greatly
from tract to tract. Figure 3a shows that identied VBs can be
arbitrarily grouped into three clusters of very hard, hard, and
medium difculty, according to the percentage of OL. Figure 3b
shows corresponding examples that were reconstructed by dif-
ferent tractography techniques. All submissions had difculties
reconstructing the smallest tracts, that is, the anterior (CA) and
posterior commissures (CP) that have a cross-sectional diameter
of no more than 2 mm, or one or two voxels (very hard, 0% <=
OL <10%). A family of hard bundles was partly recovered (10%
<= OL <50%). Bundles of medium difculty were the corpus
callosum (CC), inferior longitudinal fasciculus (ILF), superior
longitudinal fasciculus (SLF), and uncinate fasciculus (UF) with
an average of more than 50% volumetric recovery (50% <= OL
<= 100%). A Pearson product-moment correlation coefcient
was computed to assess the relationship between OL and OR (r=
0.88, p<108), indicating a direct link between the probability of
reconstructing a greater portion of a tract (OL) and generating
artefactual trajectories (OR).
Figure 4shows that on average 21 out of 25 VBs (median 23)
were identied by the participating teams with only four teams
submitting tractograms that contained an OL of more than 60%.
No submission contained all 25 VBs, but 10 submissions (10.4%)
recovered 24 VBs, and 69 submissions (71.9%) detected 23 or
more VBs (Fig. 5a). However, tractography pipelines clearly need
to improve their recovery of the full spatial extent of bundles: the
mean value of bundle volume overlap (OL) across all submissions
was 36 ±16%, with an average overreach (OR) of 29 ±26%
(Fig. 4c). At the level of individual streamlines, an average of
54 ±23% connections were valid (Fig. 4a).
Tractograms contained more invalid than valid bundles.Across
submissions, 36 ±17% of the reconstructed individual streamlines
connected regions that were not actually connected. The fraction of
streamlines not connecting any endpoints was 10 ±15%. Even
though not part of the ground truth, these streamlines often occur
in dense, structured, and coherent bundles. Submitted tractograms
Team number
Pipeline configuration of teams
Lab location of teams
12345 6 789
10 11 12 13 1415 16 17 18 19 20
20 Organisers10 11 12 13 14 15 16 17 18 19
Fig. 2 Summary of teams and tractography pipeline setups. aLocation of the teamsafliated labs. bConguration of the different pipelines used for processing
(A: motion correction, B: rotation of b-vectors, C: distortion correction, D: spike correction, E: denoising, F: upsampling, G: diffusion model beyond diffusion
tensor imaging (DTI), H: tractography beyond deterministic, I: anatomical priors, J: streamline ltering, K: advanced streamline ltering, L: streamline clustering)
NATURE COMMUNICATIONS | DOI: 10.1038/s41467-017-01285-x ARTICLE
NATURE COMMUNICATIONS |8: 1349 |DOI: 10.1038/s41467-017-01285-x | 3
Content courtesy of Springer Nature, terms of use apply. Rights reserved
contained an average of 88 ±58 IBs, which is more than four times
the amount of VBs they contained on average (Fig. 4b). This
demonstrates the inability of current state-of-the-art tractography
algorithms to control for false positives. Forty-one of these IBs
occurred in the majority of submissions (Fig. 5, Supplementary
Fig. 3). Overall average precision on the bundle level was 23 ±9%
(recall 85 ±15%, specicity 93 ±5%). Submissions with at least 23
VBsshowednofewerthan37IBs(mean88±39, n=69). Sub-
missions with 23 or more VBs and a volumetric bundle overlap of
>50% identied between 99 and 204 IBs.
The bundles illustrated in Fig. 5b were systematically found by
8195% of submissions without being part of the ground truth.
Interestingly, several of these invalid streamline clusters exhibited
similarities in anatomical location to bundles known or
Ground truth DTI DET HARDI DET
OR-R Found by N
Very hard
0 102030405060
Overreach [%]
Overlap [%]
100/0% (OL/OR) 53/44% 85/185% 90/263%
100/0% 37/30% 52/67% 79/95%
100/0% 26/6% 48/53% 89/104%
Hard Hard MediumHard
Ver y
Not found
0/0% 15/21% 2/1%
42/24% 83/68%
Fig. 3 Tractography identies most of the ground truth bundles, but not their full extent. aOverview of scores reached for different bundles in ground truth.
Average overlap (OL) and average overreach (OR) scores for the submissions (red: very hard, green: hard, blue: medium, for abbreviations see Fig. 1). b
Representative bundles for DTI deterministic (DET) tracking come from submission 6/team 20, high angular resolution diffusion imaging (HARDI)
deterministic tracking from submission 0/team 9, and HARDI probabilistic (PROBA) tracking from submission 2/team 12 (see Supplementary Note 5for a
discussion of these submissions). The rst column shows ground truth VBs for reference. The reported OL and OR scores correspond to the highest OL
score reached within the respective class of algorithms
ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/s41467-017-01285-x
4NATURE COMMUNICATIONS |8: 1349 |DOI: 10.1038/s41467-017-01285-x |
Content courtesy of Springer Nature, terms of use apply. Rights reserved
previously debated in tractography literature, such as the frontal
aslant tract (FAT)42, the arcuate fasciculus (AF)43, the inferior
frontal occipital fasciculus (IFOF)44, the middle longitudinal
fasciculus (MdLF)45, the extreme capsule fasciculus46, the super-
ior fronto-occipital fasciculus (SFOF)44,47, and the vertical
occipital fasciculus (VOF)48. These ndings suggest that evidence
for the existence of tracts should not be taken solely from
tractography at its current state but complemented by other
anatomical and electrophysiological methods.
Higher image quality may improve tractography validity.To
conrm that our ndings revealed fundamental properties of
tractography itself and are not related to effects of our specic
phantom simulation process, we ran two independent imple-
mentations of deterministic streamline tractography (Supple-
mentary Note 2) directly on the ground truth eld of ber
orientations (Fig. 6), that is, without using the diffusion-weighted
data at all. This experiment was repeated for multiple resolutions
(2, 1.75, 1.5, 1.25, 1.0, 0.75, and 0.5 mm). This setup was, thus,
independent of image quality, artifacts, and many other inu-
ences from specic pipeline congurations and the phantom
generation process. Based on the ground truth orientations, the
tractography pipelines achieved overlap scores (76 ±6%) that
were previously unreached at similar levels of overreach
(29 ±8%). VC ratios were between 71 and 82%. However, the
tractograms still contained 102 ±24 IBs (minimum 73).
Methodological innovation may improve tractography validity.
Our results show that the geometry of many junctions in the
simulated data set is too complex to be resolved by current trac-
tography algorithms, even when given a perfect ground truth eld
of orientations. Thus, the problems seem to relate to essential
ambiguities in the directional information (Fig. 7). They persisted
in supplementary experiments performed to test the potential of
currently available anatomical constraints and global tractography
approaches (Supplementary Note 2), in which none of the addi-
tionally ran methods surpassed the challenge submissions in
bundle detection performance (Supplementary Fig. 4).
We further investigated the ambiguities tractography encounters
in the synthetic phantom as well as in an in vivo data set. In the
temporal lobe, for example, multiple bundles overlap and clearly
outnumber the count of ber orientations in most of the voxels. As
illustrated in Fig. 8,singleber directions in the diffusion signal
regularly represent multiple bundles (see also Supplementary
Movie 1). Such funnels embody hard bottlenecks for tractography,
leading to massive combinatorial possibilities of plausible cong-
urations for connecting the associated bundle endpoints as sketched
in Figs. 7cand8c. Consequently, for the real data set as well as the
synthetic phantom, dozens of structured and coherent bundles pass
through this bottleneck, exhibiting similar ber counts (cf.
Supplementary Figs. 5,6) and a wide range of anatomically
reasonable geometries as illustrated in Supplementary Movie 2.A
tractogram based on real HCP data exhibits a whole family of
theoretically plausible bundles going through the temporal lobe
bottleneck even though, locally, the diffusion signal often shows
only one ber direction (cf. Fig. 8d). Methodological innovation will
be necessary to resolve these issues and better exploit additional
information sources that complement the local orientation elds
estimated from DWI.
Statistical analysis of processing steps.Effectsofthemethodo-
logical setup of the different submissions on the results were
investigated in a multivariable linear mixed model and revealed the
inuence of the individual processing steps on the tractography
outcome (Table 1). The choice of tractography algorithm, as well as
the post-tracking ltering strategy and the underlying diffusion
modeling had a strong effect on overall scores, revealing a clear
tradeoff between sensitivity and specicity (Supplementary Note 3).
Manual editing of tractograms following anatomical priors had a
Team number
12345 6 789 10
Valid connections
Ratio [%]
11 12 13 1415 16 17 18 19 20
Bundle detection
Number invalid bundles (IB)
Number valid bundles (VB)
32 64 128 256
Bundle overreach [%]
Bundle volume
Bundle overlap [%]
60 80 160140120100
Fig. 4 Between-group differences in tractography reconstructions of VBs and IBs. Overview of the scores reached by the different teams as apercentage of
streamlines connecting valid regions, bnumber of detected VBs and IBs (data points are jittered to improve legibility), and cvolume overlap (OL) and overreach
(OR) scores averaged over bundles. Black arrows mark submissions used in the following gures (see Supplementary Note 5for discussion)
NATURE COMMUNICATIONS | DOI: 10.1038/s41467-017-01285-x ARTICLE
NATURE COMMUNICATIONS |8: 1349 |DOI: 10.1038/s41467-017-01285-x | 5
Content courtesy of Springer Nature, terms of use apply. Rights reserved
negative impact on the number of VBs identied (mean effect: 3.8
±2.6 bundles) and on the bundle overlap (mean effect: 15 ±11%).
However, such techniques showed a positive impact on the average
bundle overreach (mean effect: 16 ±9%). Notably, Team 3 post-
processed the tractograms using clustering, reaching 92% validly
connecting streamlines keeping only the larger clusters.
We assessed current state-of-the-art ber tractography approa-
ches using a ground truth data set of white matter tracts and
connectivity that is representative of the challenges that may
occur in human in vivo brain imaging. Advanced tractography
strategies in combination with current diffusion modeling
techniques successfully recovered most VBs, covering up to 77%
of their volumetric extent. This result demonstrates the capability
of current methods and teams to adequately handle numerous
artifacts in DWI and overcome local crossing situations during
tract reconstruction. However, tractography also produced thick
and dense bundles of plausible looking streamlines in locations
where such streamlines did not actually exist. When focusing on
the 64 bundles that were systematically recovered by the majority
of submissions, 64% of them were in fact absent from the ground
truth. Current tractography pipelines, and even tracking of the
ground truth ber orientations on high-resolution images, pro-
duce substantial amounts of false-positive bundles. The employed
simulation-based approach cannot quantify the effects related to
in vivo connectivity in an absolute sense; that is, our results do
85% of submissions
81% of submissions
95% of submissions
81% of submissions
95% of submissions
88% of submissions
Found by 1
Found by 24
Found by 48
Found by 72
Found by 1
Found by 24
Found by 48
Found by 72
Red: valid
Blue: invalid
Fig. 5 Overview of VBs and IBs and examples of invalid streamline clusters. aEach entry in the connectivity matrix indicates the number of submissions
that have identied the respective bundle. The two rows and columns of each bundle represent the head-endpoint and tail-endpoint regions. The
connectivity matrix indicates a high number of existing tracts that were identied by most submissions (red). It also indicates systematic artefactual
reconstructions across teams (blue). bExamples of IBs that have been consistently identied by more than 80% of the submissions, but do not exist in the
ground truth data set. The AF, for example, was generated from ILF and SLF crossing streamlines, whereas the IFOF was generated from by crossing ILF and
UF streamlines. The MdLF, FAT, SFOF, and VOF were other examples of highly represented IBs
ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/s41467-017-01285-x
6NATURE COMMUNICATIONS |8: 1349 |DOI: 10.1038/s41467-017-01285-x |
Content courtesy of Springer Nature, terms of use apply. Rights reserved
not mean that anyone who is doing tractography should expect
the reported VB-to-IB and coverage-to-overreach ratios. How-
ever, the presented ndings do expose the degree of ambiguity
associated with whole-brain tractography and show that the
computational problem of tractography goes far beyond the local
reconstruction of ber directions1,7and issues of data quality.
Our ndings, therefore, present a core and open challenge for the
eld of tractography and connectivity mapping in general.
Previous studies have reported high invalid-connection ratios
under simplied conditions26,38 (, and some
of the underlying ambiguities in tractography have been discussed
using schematic representations and theoretical arguments1,7,8,37.
Regions of white matter bottlenecks have been discussed in the
past35 and have been highlighted as critical with respect to tracto-
graphic ndings36. The current results reveal the consequences of
such limitations under more complex conditions as might be found
in human brain studies in vivo, addressing important questions that
previously remained speculative. The ndings were derived from a
brain-like geometry that encompasses some of the major known
long-range connections and covers 71% of the white matter. Future
versions of the phantom are planned to include additional bundles
such as the middle and inferior temporal projections of the AF, the
MdLF, and the IFOF, as well as smaller U-bers, medial forebrain
bers, deep nuclei, and connections between them. In addition,
more advanced diffusion modeling methods will allow generating
even more realistic DWI signals, potentially simulated at increased
spatial and q-space resolutions49.
These developments, however, will not resolve the fundamental
ambiguities which tractography faces and thus will only have a
limited effect on the main ndings of our study. We showed that
false-positive bundles occur at similar rates even when using the
maximal angular precision of the signal, that is, using ground
truth orientations. These ndings conrm those shown in pre-
vious studies5and relate to the fundamental problem formulation
in tractography: inferring connectivity from local orientation
elds. Increasing the anatomic complexity of the phantom by
adding more bundles most likely will even lead to further
increased false-positive rates. The construction process of the
current phantom resembles a potential limitation, since it
involves tractography itself and thus raises self-validation issues.
This aspect should be considered in direct method comparisons
as there may exist a possible bias toward algorithms that are
similar to the algorithm used for phantom generation. This
caveat, however, has only a very limited effect on our general
ndings. It can be expected that the identied limitations of
tractography will become even more pronounced in phantoms of
higher anatomic complexity that might be achievable by involving
independent methods such as polarized light imaging50.In
summary, our observations conrm the fundamental ill-posed
nature of the computational problem that current tractography
approaches strive to solve.
Accordingly, substantial methodological innovations will be
necessary to resolve the problem of IBs. Several directions of
current research might improve the specicity of tractography.
Streamline ltering techniques can optimize the signal prediction
error in order to reduce tractography biases14,16,51. They are part
of the more general trend to integrate non-local information, as
well as advanced diffusion microstructure modeling that goes
beyond the raw directional vectors5258. Recent advances in
machine-learning-driven tractography also show great potential
in improving the specicity of tractograms59,60. Future versions
of our phantom will be generated with multiple b-values, better
signal-to-noise ratio (SNR), and fewer artifacts to further
encourage research in these directions.
In addition, tractography should increasingly employ reliable
anatomical priors from ex vivo histology, high-resolution post-
mortem DWI61, or complementary electrophysiology for optimal
guidance. While manual or automated clean-up of streamlines
Team number
Valid connections
Ratio [%]
11 12 13 1415 16 17 18 19 20 GT1GT2
Bundle detection
Number of invalid bundles
Number of valid bundles
32 64 128 256
Bundle overreach [%]
Bundle volume
Bundle overlap [%]
Team markers
60 80 16014012010040
GT1 1.25 mm
GT1 1.5 mm
GT1 1.75 mm
GT1 2.0 mm
GT2 0.5 mm
GT2 0.75 mm
GT2 1.0 mm
GT2 1.25 mm
GT2 1.5 mm
GT2 1.75 mm
GT2 2.0 mm
GT1 0.5 mm
GT1 0.75 mm
GT1 1.0 mm
Fig. 6 Tractography on ground truth directions with no noise still affected by IB problem. We applied deterministic tractography directly to the ground truth
vector eld with multiple resolutions (2, 1.75, 1.5, 1.25, 1.0, 0.75, and 0.5 mm). Two independent implementations of deterministic tractography methods
were used to obtain the results (GT
and GT
, cf. Supplementary Note 2). aPercentage of streamlines connecting valid regions. bNumber of detected VBs
and IBs (data points are jittered to improve legibility). cVolume overlap and overreach scores averaged over bundles
NATURE COMMUNICATIONS | DOI: 10.1038/s41467-017-01285-x ARTICLE
NATURE COMMUNICATIONS |8: 1349 |DOI: 10.1038/s41467-017-01285-x | 7
Content courtesy of Springer Nature, terms of use apply. Rights reserved
may help (as demonstrated by our results showing decreased
overreach at the expense of VB detection and volumetric recon-
struction), the real challenge is our limited knowledge of the
anatomy to be reconstructed. Currently, post-mortem dissection
with Klinglers method reveals the macroscopic organization of
the human brain white matter11,6264, although this method
shares some of the mentioned limitations of tractography in
complex ber congurations or near the cortex. In the future, the
community will have to gain further insights into the underlying
principles of white matter organization and increasingly learn
how to leverage such information for tractography1,65,66.
Potential advances achieved in tractography will have an
important impact on graph-analytical studies of the structural
connectome2,67. The hitherto demonstrated diagnostic or pre-
dictive capability of such analyses (e.g., in psychiatric settings)
should not let us overlook which aspects of tractography are
reliable and which are not. One of the present ndings is parti-
cularly relevant for the eld of connectomics: the traditional
metrics that require streamlines to exactly end in head or tail
regions of a bundle are far too restrictive for bundle dissection
and connectivity assessment. None of the submissions generated
exact streamlines that perfectly overlap with ground truth bundles
and dilated endpoint masks. This nding is in line with previous
reports, which found termination of tracts in the gray matter
(GM) to be inaccurate5and highlights an important limitation of
approaches that use a voxel-wise denition of parcellations on the
T1 image for selecting relevant streamlines. Future versions of our
phantom will include ground-truth parcellations of the white
matter/GM cortical band to encourage further developments for
tackling these problems and extend the evaluation method to
apply to graph theory metrics.
Despite any limitations, DWI is currently the only tool to map
short and long-range structural brain connectivity in vivo and is
essential for comparing brains, detecting differences, and simu-
lating brain activity39. Our ndings should foster the develop-
ment of novel tractography methods that are carefully evaluated
using the present approach. The most important goal for the next
generation of tractography algorithms is an improved ability to
reconstruct the full spatial extent of existing tracts while better
controlling for false-positive connections. A tighter integration of
anatomical priors, advanced diffusion microstructure modeling,
and multi-modality imaging should help to resolve ambiguities in
the signal and overcome current limitations of tractography57,58.
Fundamentally, there is an urgent need for methodological
innovation in tractography in order to advance our knowledge of
human white matter anatomy and build anatomically correct
human connectomes1,7.
Generation of ground truth ber bundles. The set of ground truth long-range
ber bundles was designed to cover the whole human brain and features many of
the relevant spatial congurations, such as crossing, kissing, twisting and fanning
bers, thus representing the morphology of the major known in vivo ber bundles.
The process to obtain these bundles consisted of three steps. First, a whole-brain
global tractography was performed on a high-quality in vivo diffusion-weighted
image. Then, 25 major long-range bundles were manually extracted from the
resulting tractogram. In the third step, these bundles were rened to obtain smooth
and well-dened bundles. Each of these steps is detailed in the following
We chose one of the diffusion-weighted data sets included in the Q3 data
release of the HCP39, subject 100307, to perform whole-brain global ber
tractography52,68. Among other customizations, the HCP scanners are equipped
with a set of high-end gradient coils, enabling diffusion encoding gradient strengths
of 100 mT m1. By comparison, most standard magnetic resonance scanners
feature gradient strengths of about 30 to 40 mT m1. This hardware setup allows
the acquisition of data sets featuring exceptionally high resolutions (1.25 mm
isotropic, 270 gradient directions) while maintaining an excellent SNR. All data sets
were corrected for head motion, eddy currents and susceptibility distortions and
are, in general, of very high quality6973. Detailed information regarding the
employed imaging protocols as well as the data sets themselves may be found on
Global ber tractography was performed using MITK Diffusion74 with the
following parameters: 900,000,000 iterations, a particle length of 1 mm, a particle
width of 0.1 mm, and a particle weight of 0.002. Furthermore, we repeated the
tractography six times and combined the resulting whole-brain tractograms into
one large data set consisting of over ve million streamlines. The selected
parameters provided for a very high sensitivity of the tractography method. The
specicity of the resulting tractogram was of lesser concern since the tracts of
interest were extracted manually in the second step.
Bundle segmentation was performed by an expert radiologist using manually
placed inclusion and exclusion regions of interest (ROI). We followed the concepts
introduced in ref. 40 for the ROI placement and ber extraction. Twenty-ve
(A) Voxel level(B) Local level(C) Global level
ground truth
Illustration of
imaging information
Unordered selection of
potential tract hypotheses
Fig. 7 Ambiguous correspondences between diffusion directions and ber geometry. The three illustrations at voxel, local, and global level are used as an
example to illustrate the possible ambiguities contained in the diffusion imaging information that may lead to alternative tract reconstructions. (A) The
intra-voxel crossing of bers in the hypothetical ground truth leads to ambiguous imaging information at voxel level7. (B) Similarly, the imaging
representation of local ber crossings can be explained by several other congurations7. (C) At a global level, white matter regions that are shared by
multiple bundles (so-called bottlenecks, dotted rectangles)35 can lead to many spurious tractographic reconstructions36. With only two bundles in the
hypothetical ground truth (red and yellow bundle), four potential false-positive bundles emerge. Please note that the hypothetical ground truth used in the
global-level example is anatomically incorrect as most of the callosal bers are homotopically distributed (i.e., connect similar regions on both hemispheres)
ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/s41467-017-01285-x
8NATURE COMMUNICATIONS |8: 1349 |DOI: 10.1038/s41467-017-01285-x |
Content courtesy of Springer Nature, terms of use apply. Rights reserved
bundles were extracted, covering association, projection, and commissural bers
across the whole brain (Fig. 1): CC, left and right cingulum (Cg), Fornix (Fx),
anterior commissure (CA), left and right optic radiation (OR), posterior
commissure (CP), left and right inferior cerebellar peduncle (ICP), middle cerebellar
peduncle (MCP), left and right superior cerebellar peduncle (SCP), left and right
parieto-occipital pontine tract (POPT), left and right cortico-spinal tract (CST), left
and right frontopontine tracts (FPT), left and right ILF, left and right UF, and left
and right SLF. As mentioned in the Discussionsection, the IFOF, the MdLF, as
well as the middle and inferior temporal projections of the AF were not included.
After manual extraction, the individual long-range bundles were further rened
to serve as ground truth for the image simulation as also shown in Fig. 1. The
original extracted tracts featured a large number of prematurely ending bers and
the individual streamlines were not smooth. To obtain smooth tracts without
prematurely ending bers, we simulated a diffusion-weighted image from each
original tract individually using Fiberfox (www.mitk.org33). Since no complex ber
congurations, such as crossings, were present in the individual tract images and no
artifacts were simulated, it was possible to obtain very smooth and complete tracts
from these images with a simple tensor-based streamline tractography.
Supplementary Fig. 7illustrates the result of this rening procedure on the left CST.
Simulation of phantom images with brain-like geometry. The phantom
diffusion-weighted images (Supplementary Movie 3) were simulated using Fiberfox
(www.mitk.org33), which is available as open-source software. We employed a
four-compartment model of brain tissue (intra and inter-axonal), GM, and cere-
brospinal uid (CSF)33. The parameters for simulation of the four-compartment
diffusion-weighted signal were chosen to obtain representative diffusion properties
and image contrasts (compare75 for details on the models). The intra-axonal
compartment was simulated using the stick model with a T2 relaxation time of 110
ms and a diffusivity of 1.2 × 10-9 m2s1. The inter-axonal compartment was
simulated using the zeppelin model with a T2 relaxation time of 110 ms, an axial
diffusivity of 1.2 × 10-9 m2s1and a radial diffusivity of 0.3 × 10-9 m2s1. The GM
compartment was simulated using the ball model with a T2 relaxation time of 80
ms and a diffusivity of 1.0 × 10-9 m2s1. The CSF compartment was also simulated
using the ball model with a T2 relaxation time of 2500 ms and a diffusivity of 2.0 ×
10-9 m2s1.
Using Fiberfox, one set of diffusion-weighted images and one T1-weighted
image were simulated. The nal data sets as well as all les needed to perform the
simulation are available online (see Data availability).
The acquisition parameters that we report below were chosen to simulate
images that are representative for a practical (e.g., clinical) setting, specically a
510-min single shot echo-planar imaging scan with 2 mm isotropic voxels, 32
gradient directions, and a b-value of 1000 s mm2. The chosen acquisition setup
represents a typical scenario for an applied tractography study and embodies a
common denominator supported by the large majority of methods. Since
acquisitions with higher b-values, more gradient directions and fewer artifacts are
Synthetic phantom data
# Signal peaks
Invalid bundles through bottleneck (selection)
Valid bundles through bottleneck
Suggested bundles through bottleneck (selection)
Streamlines through bottleneck
# Signal peaks
# Valid bundles
# Valid bundles
In vivo HCP real data
Synthetic phantom data In vivo HCP real data
Fig. 8 Bottlenecks and the fundamental ill-posed nature of tractography. aVisualization of six ground truth bundles converging into a nearly parallel funnel
in the bottleneck region of the left temporal lobe (indicated by square region). The bundles per voxel (box # Valid bundles) clearly outnumber the peak
directions in the diffusion signal (box # Signal peaks). bVisualization of streamlines from a HCP in vivo tractogram passing through the same region. c
Exemplary IBs that have been identied by more than 50% of the submissions, showing that tractography cannot differentiate between the high amount of
plausible combinatorial possibilities connecting different endpoint regions (see Supplementary Movie 1). dAutomatically QuickBundle-clustered
streamlines from the in vivo tractogram going through the temporal ROI. The clustered bundles are illustrated in different shades of green. These clusters
represent a mixture of true-positive and false-positive bundles going through that bottleneck area of the HCP data set (see Supplementary Movie 2)
NATURE COMMUNICATIONS | DOI: 10.1038/s41467-017-01285-x ARTICLE
NATURE COMMUNICATIONS |8: 1349 |DOI: 10.1038/s41467-017-01285-x | 9
Content courtesy of Springer Nature, terms of use apply. Rights reserved
benecial for tractography, we additionally report a least upper bound tractography
performance under perfect image quality conditions using a data set that directly
contains ground truth ber orientation information at high spatial resolution with
no artifacts (Fig. 6and Supplementary Note 2).
The parameters are a matrix size of 90 × 108 × 90, echo time (TE) 108 ms, dwell
time 1 ms; T2relaxation time 50 ms. The simulation corresponded to a single-coil
acquisition with constant coil sensitivity, no partial Fourier and no parallel
imaging. Phase encoding was posterior-anterior. Two unweighted images with
posterior-anterior/anterior-posterior phase encoding were also generated.
Since Fiberfox simulates the actual k-space acquisition, it was possible to
introduce a number of common artifacts into the nal image. Complex Gaussian
noise was simulated yielding a nal SNR relative to the mean white matter baseline
signal of about 20. Ten spikes were distributed randomly throughout the image
volumes (Fig. 9a). N/2 ghosts were simulated (Fig. 9b). Distortions caused by B
eld inhomogeneities are introduced using an existing eld map measured in a real
acquisition and registered to the employed reference HCP data set (Fig. 9c). Head
motion was introduced as random rotation (±4° around z-axis) and translation
(±2 mm along x-axis) in three randomly chosen volumes. Volume 6 was rotated by
3.36° and translated by 1.74 mm, volume 12 was rotated by 1.23° and translated
by 0.72 mm, and volume 24 was rotated by 3.12° and translated by 1.55 mm.
The image with the T1-like contrast was generated at an isotropic resolution of
1 mm, an SNR of about 40 and no further artifacts as an anatomical reference.
Performance metrics and evaluation. The groups submitted sets of streamlines
(see Data availability) and a brief description of their methods which is available in
Supplementary Note 1. Potential operator-dependent errors were not taken into
account but these are likely to have contributed to the quality of the nal results.
Probabilistic tractography techniques were preprocessed with a user-dened
uncertainty threshold that each group decided independently before submission.
The Tractometer denition of a VC is extremely restrictive for current
tractography algorithms, as it requires streamlines (1) not to exit the area of the
ground truth bundle at any point and (2) to terminate exactly within the endpoint
region that is dened by the dilated ground truth ber endpoints (Supplementary
Figs. 8,9)38. Hence, we adopted an alternative denition with less stringent criteria
based on robust shape distance measures76 and clustering between streamlines77,as
detailed in Supplementary Note 4. The bundle-specic thresholds were manually
congured to account for bundle shape and proximity to other bundles. The
following distances were used, with identical distances on both sides for lateralized
bundles: 2 mm for CA and CP; 3 mm for CST and SCP; 5 mm for Cingulum; 6 mm
for Fornix, ICP, OR, and UF; 7 mm for FPT, ILF, and POPT; 10 mm for CC, MCP,
and SLF. The full script used to run this bundle recognition implementation was
based on the DIPY library78 ( and is available online
(Supplementary Software 1).
Once VCs are identied, the remaining streamlines can be classied into ICs
and non-connecting streamlines. The details of this procedure are described in
Supplementary Note 4. We clustered the remaining invalid streamlines using a
QuickBundles-based clustering algorithm77. The best matching endpoint regions
for each resulting cluster were identied by majority voting of the contained
streamlines. If multiple clusters were assigned to the same pair of regions, they were
merged. Streamlines that were not assigned to any cluster or that fell below a length
threshold were labeled as non-connecting.
On the basis of this classication of streamlines, the following metrics were
1. VC ratio: Number of VCs/total number of streamlines (percentage between 0
and 100).
2. VB: For each bundle that has at least one valid streamline associated with it,
this counter is incremented by one (integer number between 0 and 25).
3. IB: With 25 bundles in the ground truth, each having two endpoint regions,
there are 1275 possible combinations of endpoint regions. Taking the 25 VBs
out of the equation, 1250 potential IBs remain (integer number between 0 and
4. Overlap: Proportion of the voxels within the volume of a ground truth bundle
that is traversed by at least one valid streamline associated with the bundle.
This value shows how well the tractography result recovers the original
volume of the bundle (percentage between 0 and 100).
5. Overreach: Fraction of voxels outside the volume of a ground truth bundle
that is traversed by at least one valid streamline associated with the bundle
ab c
Fig. 9 Illustration of artifacts included in the synthetic data set. Exemplary illustration of the spike (a), N/2 ghost (b), and distortion artifacts (c) contained
in the nal diffusion-weighted data set. Supplementary Movie 3gives an impression of the complete synthetic data set provided
Table 1 Summary of the statistical analysis
Green cells indicate a signicant positive inuence (p<0.05) and red cells indicate a signicant negative impact (p<0.05). Numbers indicate the estimated mean effect on the metric and its standard
deviation. The rst column of the table represents the different parts of the processing pipeline that we have grouped into categories. The other columns represent the metrics: VC valid connections, VB
valid bundles, IB invalid bundles, OL overlap, OR overreach
ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/s41467-017-01285-x
10 NATURE COMMUNICATIONS |8: 1349 |DOI: 10.1038/s41467-017-01285-x |
Content courtesy of Springer Nature, terms of use apply. Rights reserved
over the total number of voxels within the ground truth bundle. This value
shows how much the VCs extend beyond the ground truth bundle volume
(percentage between 0 and 100). This value is always zero for the traditional
denition of a VC, but can be non-zero for the non-stringent criteria we
adopted in our study.
Following previously dened criteria of evaluation79, our study is mainly about
accuracy with respect to the reference, rather than reproducibility or robustness of
Statistical multi-variable analysis. Effects of the experimental settings were
investigated in a multivariable linear mixed model. The experimental variables
describing the methods used for the different submissions were included as xed
effects (Fig. 2b). The VC ratio, the VB count, the IB count, the bundle overlap
percentage, and the bundle overreach percentage were modeled as dependent
variables, each of which is used for the calculation of a separate model. The
submitting group was modeled as a random effect. The software SAS 9.2, Proc
Mixed, SAS Institute Inc., Cary, NC, USA, was used for the analysis.
Data availability. The authors declare that the data supporting the ndings of this
study are available within the paper and its Supplementary Information les. The
ISMRM 2015 Tractography Challenge data sets and the submitted tractograms are
available under and
zenodo.840086, respectively.
Received: 21 November 2016 Accepted: 1 September 2017
1. Jbabdi, S., Sotiropoulos, S. N., Haber, S. N., Van Essen, D. C. & Behrens, T. E.
Measuring macroscopic brain connections in vivo. Nat. Neurosci. 18,
15461555 (2015).
2. Bullmore, E. & Sporns, O. Complex brain networks: graph theoretical analysis
of structural and functional systems. Nat. Rev. Neurosci. 10, 186198 (2009).
3. Deco, G. & Kringelbach, M. L. Great expectations: using whole-brain
computational connectomics for understanding neuropsychiatric disorders.
Neuron 84, 892905 (2014).
4. Craddock, R. C. et al. Imaging human connectomes at the macroscale. Nat.
Methods 10, 524539 (2013).
5. Thomas, C. et al. Anatomical accuracy of brain connections derived from
diffusion MRI tractography is inherently limited. Proc. Natl Acad. Sci. USA 111,
1657416579 (2014).
6. Reveley, C. et al. Supercial white matter ber systems impede detection of
long-range cortical connections in diffusion MR tractography. Proc. Natl. Acad.
Sci. USA 112, E2820E2828 (2015).
7. Jbabdi, S. & Johansen-Berg, H. Tractography: where do we go from here? Brain
Connect. 1, 169183 (2011).
8. Jones, D. K. Challenges and limitations of quantifying brain connectivity in vivo
with diffusion MRI. Imaging Med. 2, 341355 (2010).
9. Pujol, S. et al. The DTI challenge: toward standardized evaluation of diffusion
tensor imaging tractography for neurosurgery. J. Neuroimaging 25, 875882
10. Feigl, G. C. et al. Magnetic resonance imaging diffusion tensor tractography:
evaluation of anatomic accuracy of different ber tracking software packages.
World Neurosurg. 81, 144150 (2014).
11. Martino, J. et al. Cortex-sparing ber dissection: an improved method for the
study of white matter anatomy in the human brain. J. Anat. 219, 531541
12. Wang, X. et al. Subcomponents and connectivity of the superior longitudinal
fasciculus in the human brain. Brain Struct. Funct. 221, 20752092 (2015).
13. Wakana, S. et al. Reproducibility of quantitative tractography methods applied
to cerebral white matter. Neuroimage 36, 630644 (2007).
14. Pestilli, F., Yeatman, J. D., Rokem, A., Kay, K. N. & Wandell, B. A. Evaluation
and statistical inference for human connectomes. Nat. Methods 11, 10581063
15. Neher, P. F., Descoteaux, M., Houde, J.-C., Stieltjes, B. & Maier-Hein, K. H.
Strengths and weaknesses of state of the art ber tractography pipelinesa
comprehensive in-vivo and phantom evaluation study using Tractometer. Med.
Image Anal. 26, 287305 (2015).
16. Daducci, A., Dal Palù, A., Lemkaddem, A. & Thiran, J.-P. COMMIT: convex
optimization modeling for microstructure informed tractography. IEEE Trans.
Med. Imaging 34, 246257 (2015).
17. Dyrby, T. B. et al. Validation of in vitro probabilistic tractography. Neuroimage
37, 12671277 (2007).
18. Campbell, J. S., Siddiqi, K., Rymar, V. V., Sadikot, A. F. & Pike, G. B. Flow-
based ber tracking with diffusion tensor and q-ball data: validation and
comparison to principal diffusion direction techniques. Neuroimage 27,
725736 (2005).
19. Dauguet, J. et al. Comparison of ber tracts derived from in-vivo DTI
tractography with 3D histological neural tract tracer reconstruction on a
macaque brain. Neuroimage 37, 530538 (2007).
20. Schmahmann, J. D. et al. Association bre pathways of the brain: parallel
observations from diffusion spectrum imaging and autoradiography. Brain 130,
630653 (2007).
21. Seehaus, A. K. et al. Histological validation of DW-MRI tractography in human
postmortem tissue. Cereb. Cortex 23, 442450 (2013).
22. Knösche, T. R., Anwander, A., Liptrot, M. & Dyrby, T. B. Validation of
tractography: comparison with manganese tracing. Hum. Brain Mapp. 36,
41164134 (2015).
23. Donahue, C. J. et al. Using diffusion tractography to predict cortical connection
strength and distance: a quantitative comparison with tracers in the monkey. J.
Neurosci. 36, 67586770 (2016).
24. Bach, M., Maier-Hein (ne Fritzsche), K. H., Stieltjes, B. & Laun, F. B.
Investigation of resolution effects using a specialized diffusion tensor phantom.
Magn. Reson. Med. 71, 11081116 (2013).
25. Fieremans, E. et al. The design of anisotropic diffusion phantoms for the
validation of diffusion weighted magnetic resonance imaging. Phys. Med. Biol.
53, 54055421 (2008).
26. Fillard, P. et al. Quantitative evaluation of 10 tractography algorithms on a
realistic diffusion MR phantom. Neuroimage 56, 220234 (2011).
27. Maier-Hein (ne Fritzsche), K. H., Laun, F. B., Meinzer, H.-P. & Stieltjes, B.
Opportunities and pitfalls in the quantication of ber integrity: what can we
gain from Q-ball imaging? Neuroimage 51, 242251 (2010).
28. Moussavi-Biugui, A., Stieltjes, B., Fritzsche, K., Semmler, W. & Laun, F. B.
Novel spherical phantoms for Q-ball imaging under in vivo conditions. Magn.
Reson. Med. 65, 190194 (2011).
29. Poupon, C. et al. New diffusion phantoms dedicated to the study and validation
of high-angular-resolution diffusion imaging (HARDI) models. Magn. Reson.
Med. 60, 12761283 (2008).
30. Pullens, P., Roebroeck, A. & Goebel, R. Ground truth hardware phantoms for
validation of diffusion-weighted MRI applications. J. Magn. Reson. Imaging 32,
482488 (2010).
31. Close, T. G. et al. A software tool to generate simulated white matter structuresfor
the assessment of bre-tracking algorithms. Neuroimage 47, 12881300 (2009).
32. Leemans, A., Sijbers, J., Verhoye, M., Van der Linden, A. & Van Dyck, D.
Mathematical framework for simulating diffusion tensor MR neural ber
bundles. Magn. Reson. Med. 53, 944953 (2005).
33. Neher, P. F., Laun, F. B., Stieltjes, B. & Maier-Hein, K. H. Fiberfox: facilitating
the creation of realistic white matter software phantoms. Magn. Reson. Med. 72,
14601470 (2014).
34. Perrone, D. et al. D-BRAIN: anatomically accurate simulated diffusion MRI
brain data. PLoS ONE 11, e0149778 (2016).
35. Mangin, J.-F., Regis, J. & Frouin, V. Shape bottlenecks and conservative ow
systems. In Proceedings of the 1996 Workshop on Mathematical Methods in
Biomedical Image Analysis 131-138 (IEEE Computer Society, 1996).
36. Guevara, P. et al. Robust clustering of massive tractography datasets.
Neuroimage 54, 19751993 (2011).
37. Basser, P. J. Fiber-tractography via diffusion tensor MRI. in Proc. International
Society for Magnetic Resonance in Medicine 1226 (1998).
38. Cote, M. A. et al. Tractometer: towards validation of tractography pipelines.
Med. Image Anal. 17, 844857 (2013).
39. Glasser, M. F. et al. The Human Connectome Projects neuroimaging approach.
Nat. Neurosci. 19, 11751187 (2016).
40. Stieltjes, B., Brunner, R. M., Maier-Hein (ne Fritzsche), K. H. & Laun, F. B. Diffusion
Tensor Imaging: Introduction and Atlas. (Springer, Berlin Heidelberg, 2013).
41. Catani, M. & Schotten, M. T. de. Atlas of Human Brain Connections. (OUP,
Oxford, 2012).
42. Catani, M. et al. A novel frontal pathway underlies verbal uency in primary
progressive aphasia. Brain J. Neurol. 136, 26192628 (2013).
43. de Schotten, M. T. et al. A lateralized brain network for visuospatial attention.
Nat. Neurosci. 14, 12451246 (2011).
44. Forkel, S. J. et al. The anatomy of fronto-occipital connections from early blunt
dissections to contemporary tractography. Cortex 56,7384 (2014).
45. Makris, N. et al. Human middle longitudinal fascicle: variations in patterns of
anatomical connections. Brain Struct. Funct. 218, 951968 (2013).
46. Mars, R. B. et al. The extreme capsule ber complex in humans and macaque
monkeys: a comparative diffusion MRI tractography study. Brain Struct. Funct.
221, 40594071 (2015).
47. Meola, A., Comert, A., Yeh, F.-C., Stefaneanu, L. & Fernandez-Miranda, J. C.
The controversial existence of the human superior fronto-occipital fasciculus:
connectome-based tractographic study with microdissection validation. Hum.
Brain Mapp. 36, 49644971 (2015).
NATURE COMMUNICATIONS | DOI: 10.1038/s41467-017-01285-x ARTICLE
NATURE COMMUNICATIONS |8: 1349 |DOI: 10.1038/s41467-017-01285-x | 11
Content courtesy of Springer Nature, terms of use apply. Rights reserved
48. Yeatman, J. D. et al. The vertical occipital fasciculus: a century of controversy
resolved by in vivo measurements. Proc. Natl Acad. Sci. USA 111, E5214E5223
49. Maier-Hein, K. H. et al. Tractography challenge ISMRM 2015 high-resolution
data. Zenodo (2017).
50. Larsen, L., Grifn, L. D., Graessel, D., Witte, O. W. & Axer, H. Polarized light
imaging of white matter architecture. Microsc. Res. Tech. 70, 851863 (2007).
51. Smith, R. E., Tournier, J.-D., Calamante, F. & Connelly, A. The effects of SIFT
on the reproducibility and biological accuracy of the structural connectome.
Neuroimage 104, 253265 (2015).
52. Neher, P. F. et al. MITK global tractography. In SPIE Medical Imaging: Image
Processing (Eds. Haynor, D. R. & Ourselin, S.) 83144D (SPIE, 2012).
53. Mangin, J.-F. et al. Toward global tractography. Neuroimage 80, 290296
54. Jbabdi, S., Woolrich, M. W., Andersson, J. L. R. & Behrens, T. E. J. A Bayesian
framework for global tractography. Neuroimage 37, 116129 (2007).
55. Christiaens, D. et al. Global tractography of multi-shell diffusion-weighted
imaging data using a multi-tissue model. Neuroimage 123,89101 (2015).
56. Reisert, M., Kiselev, V. G., Dihtal, B., Kellner, E. & Novikov, D. S. MesoFT:
unifying diffusion modelling and ber tracking. Med. Image Comput. Comput.
Assist. Interv. 17, 201208 (2014).
57. Girard, G., Fick, R., Descoteaux, M., Deriche, R. & Wassermann, D. AxTract:
microstructure-driven tractography based on the ensemble average propagator.
Inf. Process. Med. Imaging Proc. Conf.24, 675686 (2015).
58. Daducci, A., Dal Palú, A., Descoteaux, M. & Thiran, J.-P. Microstructure
informed Tractography: pitfalls and open challenges. Front. Neurosci. 10, 247
59. Neher, P. F., Götz, M., Norajitra, T., Weber, C. & Maier-Hein, K. H. A machine
learning based approach to ber tractography using classier voting. in
International Conference on Medical Image Computing and Computer-Assisted
Intervention (Eds. Navab, N., Hornegger, J., Wells, W. & Frangi, A.) 4552
(Springer, 2015).
60. Neher, P. F., Côté, M.-A., Houde, J.-C., Descoteaux, M. & Maier-Hein, K. H.
Fiber tractography using machine learning. Neuroimage 158, 417429 (2017).
61. DellAcqua, F., Bodi, I., Slater, D., Catani, M. & Modo, M. MR diffusion
histology and micro-tractography reveal mesoscale features of the human
cerebellum. Cerebellum 12, 923931 (2013).
62. Zemmoura, I. et al. FIBRASCAN: a novel method for 3D white matter tract
reconstruction in MR space from cadaveric dissection. Neuroimage 103,
106118 (2014).
63. De Benedictis, A. et al. New insights in the homotopic and heterotopic connectivity
of the frontal portion of the human corpus callosum revealed by microdissection
and diffusion tractography. Hum. Brain Mapp. 37, 47184735 (2016).
64. Hau, J. et al. Revisiting the human uncinate fasciculus, its subcomponents and
asymmetries with stem-based tractography and microdissection validation.
Brain Struct. Funct. 222, 16451662 (2016).
65. Wedeen, V. J. et al. The geometric structure of the brain ber pathways. Science
335, 16281634 (2012).
66. Galinsky, V. L. & Frank, L. R. The lamellar structure of the brain ber
pathways. Neural Comput. 28, 25332556 (2016).
67. Sporns, O. Contributions and challenges for network models in cognitive
neuroscience. Nat. Neurosci. 17, 652660 (2014).
68. Reisert, M. et al. Global ber reconstruction becomes practical. Neuroimage 54,
955962 (2011).
69. Andersson, J. et al. A comprehensive Gaussian process framework for
correcting distortions and movements in diffusion images. in Proceedings of
International Society of Magnetic Resonance in Medicine 2426 (2012).
70. Andersson, J. L. R., Skare, S. & Ashburner, J. How to correct susceptibility
distortions in spin-echo echo-planar images: application to diffusion tensor
imaging. Neuroimage 20, 870888 (2003).
71. Fischl, B. FreeSurfer. Neuroimage 62, 774781 (2012).
72. Jenkinson, M., Beckmann, C. F., Behrens, T. E. J., Woolrich, M. W. & Smith, S.
M. FSL. Neuroimage 62, 782790 (2012).
73. Jenkinson, M., Bannister, P., Brady, M. & Smith, S. Improved optimization for
the robust and accurate linear registration and motion correction of brain
images. Neuroimage 17, 825841 (2002).
74. Maier-Hein (ne Fritzsche), K. H. et al. MITK diffusion imaging. Methods Inf.
Med. 51, 441448 (2012).
75. Panagiotaki, E. et al. Compartment models of the diffusion MR signal in brain
white matter: a taxonomy and comparison. Neuroimage 59, 22412254 (2012).
76. Garyfallidis, E., Ocegueda, O., Wassermann, D. & Descoteaux, M. Robust and
efcient linear registration of white-matter fascicles in the space of streamlines.
Neuroimage 117, 124140 (2015).
77. Garyfallidis, E., Brett, M., Correia, M. M., Williams, G. B. & Nimmo-Smith, I.
QuickBundles, a method for tractography simplication. Front. Neurosci. 6, 175
78. Garyfallidis, E. et al. Dipy, a library for the analysis of diffusion MRI data.
Front. Neuroinform. 8, 8 (2014).
79. Jannin, P. et al. Validation of medical image processing in image-guided
therapy. IEEE Trans. Med. Imaging 21, 14451449 (2002).
K.H.M.-H. was supported by the German Research Foundation (DFG), grants MA
6340/10-1, MA 6340/12-1. M.D. was supported by the NSERC Discovery Grant program
as well as the institutional Université de Sherbrooke Research Chair in Neuroinformatics.
C.M.W.T. was supported by a grant (No. 612.001.104) from the Physical Sciences
division of the Netherlands Organization for Scientic Research (NWO). The research of
H.Y.M., S.D., S.S., A.M.H., and A.L. was supported by VIDI grant 639.072.411 from
NWO. The research of F.G. was funded by the Chinese Scholarship Council (CSC).
M.Ch. was supported by the Alexander Graham Bell Canada Graduate Scholarships-
Doctoral Program (CGS-D3) from the Natural Sciences and Engineering Research
Council of Canada (NSERC). M.C. was supported by the Investigator Award No. 103759/
Z/14/Z from the Wellcome Trust. C.C.H. was supported by DFG SFB grants 936/A1, Z3
and TRR 169/A2. The research of J.-P.T., D.R., M.B., A.A., A.L., and A.D. was supported
by the Center for Biomedical Imaging (CIBM) of the Geneva-Lausanne Universities and
the EPFL, as well as the foundations Leenaards and Louis-Jeantet, and by the Swiss
National Science Foundation grants 205321_144529 and 31003A_157063. W.E.R. was
supported by CA90246 from National Cancer Institute. The research of Y.F., C.G., Y.W.,
J.M., H.R., Q.L., and C.-F.W. was supported by grant 61379020 from National Nature
Science Foundation of China. C.-F.W. was supported by NIH grants P41EB015902 and
Author contributions
K.H.M.-H., M.D., and J.-C.H. performed the data analysis and wrote the paper with
input from all authors. P.F.N. and B.S. designed the phantom. P.F.N. and J.-C.H. sup-
ported the data analysis and J.-C.H. handled the Tractometer scoring and evaluation
metrics proposed. M.-A.C. and E.G. developed the clustering and bundle recognition
algorithm for the relaxed scoring system. K.H.M.-H., P.F.N., J.-C.H., E.C., A.D., T.D.,
B.S., and M.D. coordinated the tractography challenge at the International Society for
Magnetic Resonance in Medicine (ISMRM) 2015 Diffusion Study Group meeting.
T.H.-L. set up the multivariable statistical model. P.F.N. wrote parts of the Online
Methods. L.P. and C.C.H. were mentors in the discussion of the paper and neuroana-
tomical, as well as neuroscientic context. Submissions were made by the following
teams: J.Z. team 1; M.Ch. and C.M.W.T. team 2; F.-C.Y. team 3; Y.-C.L. team 4; Q.J. team
5; D.Q.C. team 6; Y.F., C.G., Y.W., J.M., H.R., Q.L., and C.-F.W. team 7; S.D.-G., J.O.O.
G., M.P., S.S.-J., and G.G. team 8; S.S.-J., F.R., and J.S. team 9; C.M.W.T., F.G., H.Y.M., S.
D., M.F., A.M.H., and A.L. team 10; S.S.-J., G.G., and F.R. team 11; J.O.O.G., M.P., G.G.,
and F.R. team 12; A.B., B.P., C.B., M.D., S.B., and J.D. team 13; A.S., R.V., A.C., A.Q., and
J.Y. team 14; A.R.K., W.H., and S.A. team 15; D.R., M.B., A.A., O.E., A.L., and J.-P.T.
team 16; D.R., M.B., A.A., O.E., A.L., and J.-P.T. team 17; H.E.C., B.L.O., B.M., and M.S.
N. team 18; F.P., G.P., J.E.V.-R., J.G., and P.M.T. team 19; F.D.S.R., P.L.L., L.M.L., R.B.,
and F.D.A. team 20.
Additional information
Supplementary Information accompanies this paper at doi:10.1038/s41467-017-01285-x.
Competing interests: The authors declare no competing nancial interests.
Reprints and permission information is available online at
Publisher's note: Springer Nature remains neutral with regard to jurisdictional claims in
published maps and institutional afliations.
Open Access This article is licensed under a Creative Commons
Attribution 4.0 International License, which permits use, sharing,
adaptation, distribution and reproduction in any medium or format, as long as you give
appropriate credit to the original author(s) and the source, provide a link to the Creative
Commons license, and indicate if changes were made. The images or other third party
material in this article are included in the articles Creative Commons license, unless
indicated otherwise in a credit line to the material. If material is not included in the
articles Creative Commons license and your intended use is not permitted by statutory
regulation or exceeds the permitted use, you will need to obtain permission directly from
the copyright holder. To view a copy of this license, visit
© The Author(s) 2017
ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/s41467-017-01285-x
12 NATURE COMMUNICATIONS |8: 1349 |DOI: 10.1038/s41467-017-01285-x |
Content courtesy of Springer Nature, terms of use apply. Rights reserved
Klaus H. Maier-Hein
, Peter F. Neher
, Jean-Christophe Houde
, Marc-Alexandre Côté
Eleftherios Garyfallidis
, Jidan Zhong
, Maxime Chamberland
, Fang-Cheng Yeh
, Ying-Chia Lin
, Qing Ji
Wilburn E. Reddick
, John O. Glass
, David Qixiang Chen
, Yuanjing Feng
, Chengfeng Gao
Jieyan Ma
, Renjie He
, Qiang Li
, Carl-Fredrik Westin
, Samuel Deslauriers-Gauthier
J.Omar Ocegueda González
, Michael Paquette
, Samuel St-Jean
, Gabriel Girard
, François Rheault
Jasmeen Sidhu
, Chantal M.W. Tax
, Fenghua Guo
, Hamed Y. Mesri
, Szabolcs Dávid
Martijn Froeling
, Anneriet M. Heemskerk
, Alexander Leemans
, Arnaud Boré
, Basile Pinsard
Christophe Bedetti
, Matthieu Desrosiers
, Simona Brambati
, Julien Doyon
, Alessia Sarica
Roberta Vasta
, Antonio Cerasa
, Aldo Quattrone
, Jason Yeatman
, Ali R. Khan
Wes Hodges
, Simon Alexander
, David Romascano
, Muhamed Barakovic
, Anna Auría
Oscar Esteban
, Alia Lemkaddem
, Jean-Philippe Thiran
, H.Ertan Cetingul
, Benjamin L. Odry
Boris Mailhe
, Mariappan S. Nadar
, Fabrizio Pizzagalli
, Gautam Prasad
, Julio E. Villalon-Reina
Justin Galvis
, Paul M. Thompson
, Francisco De Santiago Requejo
, Pedro Luque Laguna
Luis Miguel Lacerda
, Rachel Barrett
, Flavio DellAcqua
, Marco Catani
, Laurent Petit
Emmanuel Caruyer
, Alessandro Daducci
, Tim B. Dyrby
, Tim Holland-Letz
, Claus C. Hilgetag
Bram Stieltjes
& Maxime Descoteaux
Division of Medical Image Computing, German Cancer Research Center (DKFZ), Heidelberg, 69120, Germany.
Sherbrooke Connectivity Imaging
Lab (SCIL), Université de Sherbrooke, Sherbrooke, QC J1K 0A5 QC, Canada.
Department of Intelligent Systems Engineering, School of Informatics
and Computing, Indiana University, Bloomington, IN 47408, USA.
Krembil Research Institute, University Health Network, Toronto, Canada M5G
Department of Neurological Surgery, University of Pittsburgh School of Medicine, Pittsburgh, PA 15213, USA.
IMTInstitute for Advanced
Studies, Lucca, 55100, Italy.
Department of Diagnostic Imaging, St. Jude Childrens Research Hospital, Memphis, TN 38105, USA.
University of
Toronto Institute of Medical Science, Toronto, Canada M5S 1A8.
Institute of Information Processing and Automation, Zhejiang University of
Technology, Hangzhou, 310023 Zhejiang, China.
United Imaging Healthcare Co., Shanghai, 201807, China.
Shanghai Advanced Research
Institute, Shanghai, 201210, China.
Laboratory of Mathematics in Imaging, Harvard Medical School, Boston, MA 02215, USA.
Center for
Research in Mathematics, Guanajuato, 36023, Mexico.
PROVIDI Lab, Image Sciences Institute, University Medical Center Utrecht, Utrecht, 3508,
The Netherlands.
Cardiff University Brain Research Imaging Centre, School of Psychology, Cardiff University, Maindy Road, Cardiff, CF24 4HQ,
Department of Radiology, University Medical Center Utrecht, Utrecht, 3508, The Netherlands.
Centre de recherche institut universitaire de
geriatrie de Montreal (CRIUGM), Université de Montréal, Montreal, QC, Canada H3W 1W5.
Sorbonne Universités, UPMC Univ Paris 06, CNRS,
INSERM, Laboratoire dImagerie Biomédicale (LIB), 75013 Paris, France.
Center for Advanced Research in Sleep Medicine, Hôpital du Sacré-Coeur
de Montréal, Montreal, Canada H4J 1C5.
Neuroimaging Unit, Institute of Bioimaging and Molecular Physiology (IBFM), National Research Council
(CNR), Policlinico Magna Graecia, Germaneto, 88100 CZ, Italy.
Institute of Neurology, University Magna Graecia, Germaneto, 88100 CZ, Italy.
Institute for Learning & Brain Sciences and Department of Speech & Hearing Sciences, University of Washington, Seattle, WA 98195, USA.
Departments of Medical Biophysics & Medical Imaging, Schulich School of Medicine and Dentistry, Western University, 1151 Richmond St N,
London, ON, Canada N6A 5C1.
Synaptive Medical Inc., MaRS Discovery District, 101 College Street, Suite 200, Toronto, ON, Canada M5V 3B1.
Signal Processing Lab (LTS5), Ecole Polytechnique Federale de Lausanne, Lausanne, 1015, Switzerland.
Biomedical Image Technologies (BIT),
ETSI Telecom., U. Politécnica de Madrid and CIBER-BBN, Madrid, 28040, Spain.
Department of Radiology, University Hospital Center (CHUV)
and University of Lausanne (UNIL), Lausanne, 1011, Switzerland.
Medical Imaging Technologies, Siemens Healthcare, Princeton, NJ 08540, USA.
Imaging Genetics Center, Stevens Neuro Imaging and Informatics Institute, Keck School of Medicine of USC, Marina del Rey, CA 90033, USA.
NatBrainLab, Institute of Psychiatry, Psychology & Neuroscience, Kings College London, London, SE5 8AF, UK.
Groupe dimagerie
NeurofonctionnelleInstitut des Maladies Neurodégénératives (GIN-IMN), UMR5293 CNRS, CEA, University of Bordeaux, Bordeaux, 33000,
Centre national de la recherche scientique (CNRS), Institute for Research in IT and Random Systems (IRISA), UMR 6074 VISAGES
Project-Team, Rennes, 35042, France.
Danish Research Centre for Magnetic Resonance, Center for Functional and Diagnostic Imaging and
Research, Copenhagen University Hospital Hvidovre, Hvidovre, 2650, Denmark.
Department of Applied Mathematics and Computer Science,
Technical University of Denmark, Kongens Lyngby, 2800, Denmark.
Division of Biostatistics, German Cancer Research Center (DKFZ),
Heidelberg, 69120, Germany.
Department of Computational Neuroscience, University Medical Center Eppendorf, Hamburg, 20246, Germany.
University Hospital Basel, Radiology & Nuclear Medicine Clinic, Basel, 4031, Switzerland
NATURE COMMUNICATIONS | DOI: 10.1038/s41467-017-01285-x ARTICLE
NATURE COMMUNICATIONS |8: 1349 |DOI: 10.1038/s41467-017-01285-x | 13
Content courtesy of Springer Nature, terms of use apply. Rights reserved
Terms and Conditions
Springer Nature journal content, brought to you courtesy of Springer Nature Customer Service Center GmbH (“Springer Nature”).
Springer Nature supports a reasonable amount of sharing of research papers by authors, subscribers and authorised users (“Users”), for small-
scale personal, non-commercial use provided that all copyright, trade and service marks and other proprietary notices are maintained. By
accessing, sharing, receiving or otherwise using the Springer Nature journal content you agree to these terms of use (“Terms”). For these
purposes, Springer Nature considers academic use (by researchers and students) to be non-commercial.
These Terms are supplementary and will apply in addition to any applicable website terms and conditions, a relevant site licence or a personal
subscription. These Terms will prevail over any conflict or ambiguity with regards to the relevant terms, a site licence or a personal subscription
(to the extent of the conflict or ambiguity only). For Creative Commons-licensed articles, the terms of the Creative Commons license used will
We collect and use personal data to provide access to the Springer Nature journal content. We may also use these personal data internally within
ResearchGate and Springer Nature and as agreed share it, in an anonymised way, for purposes of tracking, analysis and reporting. We will not
otherwise disclose your personal data outside the ResearchGate or the Springer Nature group of companies unless we have your permission as
detailed in the Privacy Policy.
While Users may use the Springer Nature journal content for small scale, personal non-commercial use, it is important to note that Users may
use such content for the purpose of providing other users with access on a regular or large scale basis or as a means to circumvent access
use such content where to do so would be considered a criminal or statutory offence in any jurisdiction, or gives rise to civil liability, or is
otherwise unlawful;
falsely or misleadingly imply or suggest endorsement, approval , sponsorship, or association unless explicitly agreed to by Springer Nature in
use bots or other automated methods to access the content or redirect messages
override any security feature or exclusionary protocol; or
share the content in order to create substitute for Springer Nature products or services or a systematic database of Springer Nature journal
In line with the restriction against commercial use, Springer Nature does not permit the creation of a product or service that creates revenue,
royalties, rent or income from our content or its inclusion as part of a paid for service or for other commercial gain. Springer Nature journal
content cannot be used for inter-library loans and librarians may not upload Springer Nature journal content on a large scale into their, or any
other, institutional repository.
These terms of use are reviewed regularly and may be amended at any time. Springer Nature is not obligated to publish any information or
content on this website and may remove it or features or functionality at our sole discretion, at any time with or without notice. Springer Nature
may revoke this licence to you at any time and remove access to any copies of the Springer Nature journal content which have been saved.
To the fullest extent permitted by law, Springer Nature makes no warranties, representations or guarantees to Users, either express or implied
with respect to the Springer nature journal content and all parties disclaim and waive any implied warranties or warranties imposed by law,
including merchantability or fitness for any particular purpose.
Please note that these rights do not automatically extend to content, data or other material published by Springer Nature that may be licensed
from third parties.
If you would like to use or distribute our Springer Nature journal content to a wider audience or on a regular basis or in any other manner not
expressly permitted by these Terms, please contact Springer Nature at
... By following up with this step we were able to largely reduce the known tractography issue of false positive tracking. 85 False positive fiber tracking is a contributing factor affecting white matter parcellation reproducibility. 85 In our approach, false positive fibers in the atlas have been annotated and rejected via expert judgment. ...
... 85 False positive fiber tracking is a contributing factor affecting white matter parcellation reproducibility. 85 In our approach, false positive fibers in the atlas have been annotated and rejected via expert judgment. 83 Usage of the atlas therefore can ameliorate potential subject-specific false positive fibers that are inconsistent with respect to known neuroanatomical knowledge. ...
Background: Military service members are at increased risk for mental health issues and comorbidity with mild traumatic brain injury (mTBI) is common. Largely overlapping symptoms between conditions suggest a shared pathophysiology. The present work investigates the associations between white matter microstructure, psychological functioning, and serum neuroactive steroids that are part of the stress-response system. Methods: Diffusion-weighted brain imaging was acquired from 163 participants (with and without military affiliation) and free-water-corrected fractional anisotropy (FAT) was extracted. Associations between serum neurosteroid levels of allopregnanolone (ALLO) and pregnenolone (PREGNE), psychological functioning, and whole-brain white matter microstructure were assessed using regression models. Moderation models tested the effect of mTBI and comorbid post-traumatic stress disorder (PTSD) and mTBI on these associations. Results: ALLO is associated with whole-brain white matter FAT (β=.24, t=3.00, p= .006). This association is significantly modulated by PTSD+mTBI comorbidity (β=.01, t=3.07, p=.003) while an mTBI diagnosis alone did not significantly impact this association (p=.183). There was no significant association between PREGNE and FAT (p=.380). Importantly, lower FAT is associated with poor psychological functioning (β=-.19, t=-2.35, p=.020). Conclusion: This study provides novel insight into a potential common pathophysiological mechanism of neurosteroid dysregulation underlying the high risk for mental health issues in military service members. Further, comorbidity of PTSD and mTBI may bring the compensatory effects of the brain's stress response to their limit. Future research is needed to investigate whether neurosteroid regulation may be a promising tool for restoring brain health and improving psychological functioning.
... Community challenges have consistently proved effective in moving forward the state of the art in technology to address specific data-analysis problems by providing platforms for unbiased comparative evaluation and incentives to maximise performance on key tasks (Maier-Hein et al. (2018)). In medical image analysis, for example, such challenges have provided important benchmarks in tasks such as registration (Murphy et al. (2011)) and segmentation (Menze et al. (2014)), and revealed fundamental insights about the problem studied, for example in structural brain-connectivity mapping (Maier-Hein et al. (2017)). Previous challenges in AD include the CADDementia challenge (Bron et al. (2015)), which aimed to identify clinical diagnosis from MRI scans. ...
We present the findings of "The Alzheimer's Disease Prediction Of Longitudinal Evolution" (TADPOLE) Challenge, which compared the performance of 92 algorithms from 33 international teams at predicting the future trajectory of 219 individuals at risk of Alzheimer's disease. Challenge participants were required to make a prediction, for each month of a 5-year future time period, of three key outcomes: clinical diagnosis, Alzheimer's Disease Assessment Scale Cognitive Subdomain (ADAS-Cog13), and total volume of the ventricles. No single submission was best at predicting all three outcomes. For clinical diagnosis and ventricle volume prediction, the best algorithms strongly outperform simple baselines in predictive ability. However, for ADAS-Cog13 no single submitted prediction method was significantly better than random guessing. Two ensemble methods based on taking the mean and median over all predictions, obtained top scores on almost all tasks. Better than average performance at diagnosis prediction was generally associated with the additional inclusion of features from cerebrospinal fluid (CSF) samples and diffusion tensor imaging (DTI). On the other hand, better performance at ventricle volume prediction was associated with inclusion of summary statistics, such as patient-specific biomarker trends. The submission system remains open via the website, while code for submissions is being collated by TADPOLE SHARE: Our work suggests that current prediction algorithms are accurate for biomarkers related to clinical diagnosis and ventricle volume, opening up the possibility of cohort refinement in clinical trials for Alzheimer's disease.
... The idea of inviting different analysis teams to answer the same research question using the same data is relatively novel (Silberzahn & Uhlmann, 2015; see Aczel et al., 2021 for general guidelines); we are aware of three papers in neuroscience (Botvinik-Nezer et al., 2020;Fillard et al., 2011;Maier-Hein et al., 2017), one in microeconomics (Huntington-Klein et al., 2021), and eight in psychology, three of which pertain to cognitive modeling (Boehm et al., 2018;Dutilh et al., 2019;Starnsetal.,2019) while the remaining five are from other fields of psychology (Bastiaansen et al., 2020;S a l g a n i ke ta l . ...
The relation between religiosity and well-being is one of the most researched topics in the psychology of religion, yet the directionality and robustness of the effect remains debated. Here, we adopted a many-analysts approach to assess the robustness of this relation based on a new cross-cultural dataset (N=10,535 participants from 24 countries). We recruited 120 analysis teams to investigate (1) whether religious people self-report higher well-being, and (2) whether the relation between religiosity and self-reported well-being depends on perceived cultural norms of religion (i.e., whether it is considered normal and desirable to be religious in a given country). In a two-stage procedure, the teams first created an analysis plan and then executed their planned analysis on the data. For the first research question, all but 3 teams reported positive effect sizes with credible/confidence intervals excluding zero (median reported β=0.120). For the second research question, this was the case for 65% of the teams (median reported β=0.039). While most teams applied (multilevel) linear regression models, there was considerable variability in the choice of items used to construct the independent variables, the dependent variable, and the included covariates.
... The SC-FC match between the model and data could be improved if specific measures supporting cytoarchitectonic, transcriptomic, and higher order interactions were added to the model [Suárez et al., 2020]. However, despite these additions, there remains a fundamental problem: human brain tractography is inherently limited and does not capture gray matter tracts and fibers going through thick bundles of axons, like corpus callosum [Thomas et al., 2014, Maier-Hein et al., 2017. Hence SC typically misses interhemispheric connections and, indeed artificial inclusion of those improved match to experimental FC, see Messé et al. [2014] for similar findings. ...
Brain activity during the resting state is widely used to examine brain organization, cognition and alterations in disease states. While it is known that neuromodulation and the state of alertness impact resting-state activity, neural mechanisms behind such modulation of resting-state activity are unknown. In this work, we used a computational model to demonstrate that cholinergic input influences resting-state activity and its functional connectivity through cellular and synaptic modulation. The results from the computational model match closely with experimental work on direct cholinergic modulation of Default Mode Network (DMN) in rodents. We extend those results to the human connectome derived from diffusion-weighted MRI. In the human resting state simulations, an increase in cholinergic input resulted in a brain-wide reduction of functional connectivity. Further, selective cholinergic modulation of DMN closely captured transitions between the baseline resting state and states with suppressed DMN fluctuations associated with attention to external tasks. Our study thus provides insight into potential neural mechanisms for cholinergic neuromodulation on resting state activity and its dynamics.
Full-text available
TractoInferno is the world’s largest open-source multi-site tractography database, including both research- and clinical-like human acquisitions, aimed specifically at machine learning tractography approaches and related ML algorithms. It provides 284 samples acquired from 3 T scanners across 6 different sites. Available data includes T1-weighted images, single-shell diffusion MRI (dMRI) acquisitions, spherical harmonics fitted to the dMRI signal, fiber ODFs, and reference streamlines for 30 delineated bundles generated using 4 tractography algorithms, as well as masks needed to run tractography algorithms. Manual quality control was additionally performed at multiple steps of the pipeline. We showcase TractoInferno by benchmarking the learn2track algorithm and 5 variations of the same recurrent neural network architecture. Creating the TractoInferno database required approximately 20,000 CPU-hours of processing power, 200 man-hours of manual QC, 3,000 GPU-hours of training baseline models, and 4 Tb of storage, to produce a final database of 350 Gb. By providing a standardized training dataset and evaluation protocol, TractoInferno is an excellent tool to address common issues in machine learning tractography.
The perforant path, the white matter bundle connecting the entorhinal cortex (ERC) with the hippocampal formation, is thought to deteriorate with age-related cognitive decline. Previous investigations that have used diffusion-weighted MRI to quantify perforant path integrity in-vivo have been limited due to image resolution or have quantified the perforant path using methods susceptible to partial volume effects such as the tensor model and without respect to its 3-dimensional morphology. In this investigation, we employ the use of quantitative-anisotropy informed tractography derived from ultra-high resolution diffusion imaging (ZOOMit) to investigate the structural connectivity of medial temporal lobe (MTL) regions in an aged population (63 to 98 years old, n=51). We provide evidence that structural connectivity (graph density) within the MTL declines with age and is associated with decreased delayed verbal recall performance assessed via the Rey Auditory Verbal Learning Test (RAVLT). Additionally, we provide evidence that increased age and poorer verbal recall performance is associated with a reduction in streamlines connecting the ERC and dentate gyrus of the hippocampus. While additional research is needed to verify the anatomical accuracy of these streamlines, this investigation provides evidence that the 3-dimensional perforant path may be detectable in-vivo using clinical grade scanners in aged populations. Together this work suggests that 3-dimensional perforant path integrity and MTL structural connectivity may be additional biological markers of age-related cognitive decline.
Full-text available
Auditory processing disorder (APD) is a listening impairment that some school-aged children may experience as difficulty understanding speech in background noise despite having normal peripheral hearing. Recent resting-state functional magnetic resonance imaging (MRI) has revealed an alteration in regional, but not global, functional brain topology in children with APD. However, little is known about the brain structural organization in APD. We used diffusion MRI data to investigate the structural white matter connectome of 58 children from 8 to 14 years old diagnosed with APD (n=29) and children without hearing complaints (healthy controls, HC; n=29). We investigated the rich-club organization and structural connection differences between APD and HC groups using the network science approach. The APD group showed neither edge-based connectivity differences nor any differences in rich-club organization and connectivity strength (i.e., rich, feeder, local connections) compared to HCs. However, at the regional network level, we observed increased average path length (APL) and betweenness centrality in the right inferior parietal lobule and inferior precentral gyrus, respectively, in children with APD. HCs demonstrated a positive association between APL in the left orbital gyrus and the listening-in-spatialized-noise-sentences task, a measure of auditory processing ability. This correlation was not observed in the APD group. In line with previous functional connectome findings, the current results provide evidence for altered structural networks at a regional level in children with APD, and an association with listening performance, suggesting the involvement of multimodal deficits and a role for structure-function alteration in listening difficulties of children with APD.
Mammalian taxonomies are conventionally defined by morphological traits and genetics. How species differ in terms of neural circuits and whether inter-species differences in neural circuit organization conform to these taxonomies is unknown. The main obstacle for the comparison of neural architectures have been differences in network reconstruction techniques, yielding species-specific connectomes that are not directly comparable to one another. Here we comprehensively chart connectome organization across the mammalian phylogenetic spectrum using a common reconstruction protocol. We analyze the mammalian MRI (MaMI) data set, a database that encompasses high-resolution ex vivo structural and diffusion magnetic resonance imaging (MRI) scans of 124 species across 12 taxonomic orders and 5 superorders, collected using a unified MRI protocol. We assess similarity between species connectomes using two methods: similarity of Laplacian eigenspectra and similarity of multiscale topological features. We find greater inter-species similarities among species within the same taxonomic order, suggesting that connectome organization reflects established taxonomic relationships defined by morphology and genetics. While all connectomes retain hallmark global features and relative proportions of connection classes, inter-species variation is driven by local regional connectivity profiles. By encoding connectomes into a common frame of reference, these findings establish a foundation for investigating how neural circuits change over phylogeny, forging a link from genes to circuits to behaviour.
A comprehensive description of how neurons and entire brain regions are interconnected is fundamental for a mechanistic understanding of brain function and dysfunction. Neuroimaging has shaped the way to approaching the human brain’s connectivity on the basis of diffusion magnetic resonance imaging and tractography. At the same time, polarization, fluorescence, and electron microscopy became available, which pushed spatial resolution and sensitivity to the axonal or even to the synaptic level. New methods are mandatory to inform and constrain whole-brain tractography by regional, high-resolution connectivity data and local fiber geometry. Machine learning and simulation can provide predictions where experimental data are missing. Future interoperable atlases require new concepts, including high-resolution templates and directionality, to represent variants of tractography solutions and estimates of their accuracy.
Objective Stereo-electroencephalography (SEEG)-derived epilepsy networks are used to better understand a patient’s epilepsy; however, a unimodal approach provides an incomplete picture. We combine tractography and SEEG to determine the relationship between spike propagation and the white matter architecture and to improve our understanding of spike propagation mechanisms. Methods Probablistic tractography from diffusion imaging (dMRI) of matched subjects from the Human Connectome Project (HCP) was combined with patient-specific SEEG-derived spike propagation networks. Two regions-of-interest (ROIs) with a significant spike propagation relationship constituted a Propagation Pair. Results In 56 of 59 patients, Propagation Pairs were more often tract-connected as compared to all ROI pairs (p<0.01; d= -1.91). The degree of spike propagation between tract-connected ROIs was greater (39±21%) compared to tract-unconnected ROIs (31±18%; p<0.0001). Within the same network, ROIs receiving propagation earlier were more often tract-connected to the source (59.7%) as compared to late receivers (25.4%; p<0.0001). Conclusions Brain regions involved in spike propagation are more likely to be connected by white matter tracts. Between nodes, presence of tracts suggests a direct course of propagation, whereas the absence of tracts suggests an indirect course of propagation. Significance We demonstrate a logical and consistent relationship between spike propagation and the white matter architecture.
Full-text available
Fiber tracking algorithms yield valuable information for neurosurgery as well as automated diagnostic approaches. However, they have not yet arrived in the daily clinical practice. In this paper we present an open source integration of the global tractography algorithm proposed by Reisert into the open source Medical Imaging Interaction Toolkit (MITK) developed and maintained by the Division of Medical and Biological Informatics at the German Cancer Research Center (DKFZ). The integration of this algorithm into a standardized and open development environment like MITK enriches accessibility of tractography algorithms for the science community and is an important step towards bringing neuronal tractography closer to a clinical application. The MITK diffusion imaging application, downloadable from, combines all the steps necessary for a successful tractography: preprocessing, reconstruction of the images, the actual tracking, live monitoring of intermediate results, postprocessing and visualization of the final tracking results. This paper presents typical tracking results and demonstrates the steps for pre- and post-processing of the images.
Full-text available
One of the major limitations of diffusion MRI tractography is that the fiber tracts recovered by existing algorithms are not truly quantitative. Local techniques for estimating more quantitative features of the tissue microstructure exist, but their combination with tractography has always been considered intractable. Recent advances in local and global modeling made it possible to fill this gap and a number of promising techniques for microstructure informed tractography have been suggested, opening new and exciting perspectives for the quantification of brain connectivity. The ease-of-use of the proposed solutions made it very attractive for researchers to include such advanced methods in their analyses; however, this apparent simplicity should not hide some critical open questions raised by the complexity of these very high-dimensional problems, otherwise some fundamental issues may be pushed into the background. The aim of this article is to raise awareness in the diffusion MRI community, notably researchers working on brain connectivity, about some potential pitfalls and modeling choices that make the interpretation of the outcomes from these novel techniques rather cumbersome. Through a series of experiments on synthetic and real data, we illustrate practical situations where erroneous and severely biased conclusions may be drawn about the connectivity if these pitfalls are overlooked, like the presence of partial/missing/duplicate fibers or the critical importance of the diffusion model adopted. Microstructure informed tractography is a young but very promising technology, and by acknowledging its current limitations as done in this paper, we hope our observations will trigger further research in this direction and new ideas for truly quantitative and biologically meaningful analyses of the connectivity.
Full-text available
Despite its significant functional and clinical interest, the anatomy of the uncinate fasciculus (UF) has received little attention. It is known as a ‘hook-shaped’ fascicle connecting the frontal and anterior temporal lobes and is believed to consist of multiple subcomponents. However, the knowledge of its precise connectional anatomy in humans is lacking, and its subcomponent divisions are unclear. In the present study, we evaluate the anatomy of the UF and provide its detailed normative description in 30 healthy subjects with advanced particle-filtering tractography with anatomical priors and robustness to crossing fibers with constrained spherical deconvolution. We extracted the UF by defining its stem encompassing all streamlines that converge into a compact bundle, which consisted not only of the classic hook-shaped fibers, but also of straight horizontally oriented. We applied an automatic-clustering method to subdivide the UF bundle and revealed five subcomponents in each hemisphere with distinct connectivity profiles, including different asymmetries. A layer-by-layer microdissection of the ventral part of the external and extreme capsules using Klingler’s preparation also demonstrated five types of uncinate fibers that, according to their pattern, depth, and cortical terminations, were consistent with the diffusion-based UF subcomponents. The present results shed new light on the UF cortical terminations and its multicomponent internal organization with extended cortical connections within the frontal and temporal cortices. The different lateralization patterns we report within the UF subcomponents reconcile the conflicting asymmetry findings of the literature. Such results clarifying the UF structural anatomy lay the groundwork for more targeted investigations of its functional role, especially in semantic language processing.
We present a fiber tractography approach based on a random forest classification and voting process, guiding each step of the streamline progression by directly processing raw diffusion-weighted signal intensities. For comparison to the state-of-the-art, i.e. tractography pipelines that rely on mathematical modeling, we performed a quantitative and qualitative evaluation with multiple phantom and in vivo experiments, including a comparison to the 96 submissions of the ISMRM tractography challenge 2015. The results demonstrate the vast potential of machine learning for fiber tractography.
We present a fiber tractography approach based on a random forest classification and voting process, guiding each step of the streamline progression by directly processing raw diffusion-weighted signal intensities. For comparison to the state-of-the-art, i.e. tractography pipelines that rely on mathematical modeling, we performed a quantitative and qualitative evaluation with multiple phantom and in vivo experiments, including a comparison to the 96 submissions of the ISMRM tractography challenge 2015. The results demonstrate the vast potential of machine learning for fiber tractography.
We present a quantitative statistical analysis of pairwise crossings for all fibers obtained from whole brain tractography that confirms with high confidence that the brain grid theory (Wedeen et al., 2012a) is not supported by the evidence. The overall fiber tracts structure appears to be more consistent with small angle treelike branching of tracts rather than with near-orthogonal gridlike crossing of fiber sheets. The analysis uses our new method for high-resolution whole brain tractography that is capable of resolving fibers crossing of less than 10 degrees and correctly following a continuous angular distribution of fibers even when the individual fiber directions are not resolved. This analysis also allows us to demonstrate that the whole brain fiber pathway system is very well approximated by a lamellar vector field, providing a concise and quantitative mathematical characterization of the structural connectivity of the human brain.
Noninvasive human neuroimaging has yielded many discoveries about the brain. Numerous methodological advances have also occurred, though inertia has slowed their adoption. This paper presents an integrated approach to data acquisition, analysis and sharing that builds upon recent advances, particularly from the Human Connectome Project (HCP). The 'HCP-style' paradigm has seven core tenets: (i) collect multimodal imaging data from many subjects; (ii) acquire data at high spatial and temporal resolution; (iii) preprocess data to minimize distortions, blurring and temporal artifacts; (iv) represent data using the natural geometry of cortical and subcortical structures; (v) accurately align corresponding brain areas across subjects and studies; (vi) analyze data using neurobiologically accurate brain parcellations; and (vii) share published data via user-friendly databases. We illustrate the HCP-style paradigm using existing HCP data sets and provide guidance for future research. Widespread adoption of this paradigm should accelerate progress in understanding the brain in health and disease.