Quantitative mapping of the brain’s structural connectivity using diffusion MRI tractography: A review

In the paper

“Gyral bias in tractography is the bias towards generating streamlines that terminate in gyri rather than sulci (Schilling et al., 2018a; Wu et al., 2020b), which reduces the agreement of the initiations or terminations of fiber streamlines with known ground truth from tracer studies and histology (Aydogan et al., 2018; Budde and Annese, 2013).”

What the sources say

• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites Schilling et al., 2018a
• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites Wu et al., 2020b
• Partly supported

  The Introduction (page unavailable) reports imperfect overlap between mouse tractograms and tracer injections, supporting the general point that tractography can disagree with tracer ground truth. But the paper does not attribute that disagreement to gyral bias or preferential termination in gyri rather than sulci. The Discussion (page unavailable) describes validation using histology images as future work, not a result of this study. (LLM, conf=0.79)

  cites Aydogan et al., 2018 sourceExperimental — accuracy not yet measured

In the paper

“Lastly, deep learning techniques have been increasingly applied for anatomical tract identification, showing high potential for future work and improvements in computational speed (Chen et al., 2021; Gupta et al., 2017a; 2018; 2017b; Lam et al., 2018; Reisert et al., 2018; Wasserthal et al., 2018; 2019; Xu et al., 2019; Zhang et al., 2020a).”

What the sources say

• Supported

  The Introduction describes deep learning as a promising direction for white matter fiber clustering, notes multiple existing methods for tractography segmentation, and says this paper's approach demonstrated efficiency. The Conclusion reports efficiency advantages over other algorithms and identifies directions for further improvement. (LLM, conf=0.92)

  cites Chen et al., 2021 source
• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites Gupta et al., 2017a
• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites Lam et al., 2018

In the paper

“Manual streamline selection is considered to be the gold standard to delineate anatomical fiber tracts in tractography and has been widely used to validate other anatomical tract identification techniques (Poulin et al., 2019; Pujol et al., 2015; Xie et al., 2020).”

• Also stated in: § Anatomical tract identification · plain

  “However, manual streamline selection is still considered to be the gold standard for benchmarked comparisons and algorithm training (Kreilkamp et al., 2019; Wasserthal et al., 2018; Zhang et al., 2010).”

What the sources say

• Supported

  The paper calls expert-validated streamline annotations a medical-imaging “gold standard,” while distinguishing them from biological ground truth (Annotated Datasets and Evaluation Tools, p. 4). It also reviews annotated datasets used to train or validate tractography algorithms (p. 3) and describes manually validated bundles used to train and validate a tract-segmentation method (TractSeg, p. 9). (LLM, conf=0.87)

  cites Poulin et al., 2019 source
• Partly supported

  The Introduction (page unavailable) describes manual techniques as closest to ground truth in the absence of histological data and as a benchmark for comparing automated tractography, supporting the core of the claim. It does not establish that manual selection has been widely used to validate other tract-identification techniques; it says recent automated methods have not been extensively validated. It also does not describe using manual selections for algorithm training. (LLM, conf=0.86)

  cites Kreilkamp et al., 2019 sourceExperimental — accuracy not yet measured
• Partly supported

  The Introduction describes manual virtual dissection as the most commonly employed way to segment white matter tracts, and §3.3 compares manual dissections with TractSeg and other methods. But §4.2 says the paper’s reference tracts are not real ground truth. The paper supports manual delineation and its use in comparisons, not the unqualified description of manual selection as a gold standard. (LLM, conf=0.84)

  cites Wasserthal et al., 2018 sourceExperimental — accuracy not yet measured

In the paper

“In general, studies indicate that modern multi-fiber tractography is more sensitive when performing tractography and thus is more consistent across subjects and timepoints than traditional single-fiber DTI tractography (Bucci et al., 2013; Prčkovska et al., 2016).”

What the sources say

• Partly supported

  The Discussion (‘Quantitative validation of fiber tracking methods’; page unavailable) reports significantly better sensitivity for q-ball than DTI, and ‘Implications for preoperative mapping’ calls probabilistic q-ball the most reproducible of the four methods tested. The paper does not establish consistency across subjects or timepoints, and cautions against generalizing from its small sample. (LLM, conf=0.88)

  cites Bucci et al., 2013 sourceExperimental — accuracy not yet measured
• ?Not checked — no full text of the sourcecites Prčkovska et al., 2016 source

In the paper

“Most widely used approaches perform a linear registration (Garyfallidis et al., 2015; O’Donnell et al., 2012), while many studies have proposed to perform non-linear approaches to enable deformations of the streamlines (Benou et al., 2019; Chandio and Garyfallidis, 2020; Olivetti et al., 2016).”

What the sources say

• ?Not checked — no full text of the sourcecites Garyfallidis et al., 2015 source
• Partly supported

  The Methods (section 2.3; page unavailable) describe this paper’s tractography registration as fully affine, supporting its use as an example of linear registration. The Introduction says alignment is often performed using transformations from image-based registration, but does not establish that linear approaches are the most widely used. The nonlinear-approaches clause is attributed to other citations in the sentence, not to this paper. (LLM, conf=0.82)

  cites O’Donnell et al., 2012 sourceExperimental — accuracy not yet measured
• ?Not checked — no full text of the sourcecites Benou et al., 2019 source

In the paper

“There are generally two categories of methods: cortical-parcellation-based methods and fiber clustering methods (O’Donnell et al., 2013).”

What the sources say

• Supported

  The Introduction identifies parcellation-based and clustering methods as the two main categories, while also proposing hybrid methods as a third category. Section 2 explains that the former use cortical fiber endpoints and the latter group fiber trajectories, matching the citing sentence’s distinction. (LLM, conf=0.94)

  cites O’Donnell et al., 2013 source

In the paper

“For example, the intermodality registration between dMRI and T1-weighted images can be affected by differences in image resolutions (Malinsky et al., 2013) and echo-planar imaging (EPI) distortion in dMRI data (Albi et al., 2018).”

What the sources say

• Supported

  The Background describes registration of high-resolution T1 images with low-resolution diffusion-derived FA images and says the data must be preprocessed to ensure identical resolution and voxel size. This supports the resolution-difference point attributed to Malinsky et al.; the EPI-distortion point is attributed to Albi et al. and is not assessed here. (LLM, conf=0.92)

  cites Malinsky et al., 2013 source
• Supported

  The Albi et al. citation is attached to EPI distortion in dMRI, not to the image-resolution difference. The Introduction states that EPI produces geometric distortions and displaces anatomical structures; the Methods describe registering DWI data to anatomical images to correct that distortion. (LLM, conf=0.92)

  cites Albi et al., 2018 source

In the paper

“They group tractography streamlines according to their geometric trajectories, describing the structural connectivity according to the white matter anatomy (Avila et al., 2019; Chekir et al., 2014; Garyfallidis et al., 2018; Guevara et al., 2012; Li et al., 2010; O’Donnell and Westin, 2007; Román et al., 2017; Ros et al., 2013; Siless et al., 2018; 2020; Tunç et al., 2014; Vázquez et al., 2020; Wu”

What the sources say

• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites Avila et al., 2019
• ?Not checked — no full text of the sourcecites Chekir et al., 2014 source
• Supported

  The Introduction describes streamline-based segmentation of tractograms into known anatomical bundles. Section 2.2 explains that RecoBundles clusters streamlines using a distance between their trajectories and recognizes bundles by their shape. Together, these passages support the cited description of grouping streamlines to characterize white-matter pathways; no page numbers were available in our copy. (LLM, conf=0.87)

  cites Garyfallidis et al., 2018 source

In the paper

“Many cortical-parcellation-based methods have also used a functional cortical parcellation computed using functional MRI data (Eickhoff et al., 2005; Glasser et al., 2016; Schaefer et al., 2018; Yeo et al., 2011).”

What the sources say

• ?Not checked — no full text of the sourcecites Eickhoff et al., 2005 source
• Partly supported

  The Methods (‘Modalities for parcellation’) confirm that Glasser et al. used task and resting-state fMRI data to construct a cortical parcellation. But the parcellation was multi-modal: it also used architectural measures, so describing this source as an example of a functional parcellation omits a defining qualification. (LLM, conf=0.85)

  cites Glasser et al., 2016 sourceExperimental — accuracy not yet measured
• ?Not checked — no full text of the sourcecites Schaefer et al., 2018 source

In the paper

“The first category of quantitative measures are based on the tractography data alone, which are often used to define the connectivity strength (Jones, 2010; Smith et al., 2020a; Sotiropoulos and Zalesky, 2019; Yeh et al., 2020).”

What the sources say

• ?Not checked — no full text of the sourcecites Jones, 2010 source
• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites Smith et al., 2020a
• Supported

  In “Quantifying edges,” the paper says connection strength is most typically quantified using a function of tractography-derived streamline counts. This supports the claim that measures based on tractography data alone are used to define connectivity strength; “first category” is the citing authors’ framing. (LLM, conf=0.92)

  cites Sotiropoulos and Zalesky, 2019 source

In the paper

“The most popularly used cortical parcellation is the Freesurfer Desikan-Killiany cortical parcellation (Desikan et al., 2006; Fischl, 2012), while many other cortical parcellation schemes have also been widely used (Destrieux et al., 2010; Shattuck et al., 2008; Tzourio-Mazoyer et al., 2002).”

What the sources say

• ?Not checked — no full text of the sourcecites Desikan et al., 2006 source
• Partly supported

  The Introduction (page not provided) identifies the Desikan et al. (2006) cortical parcellation as part of FreeSurfer, and the Contributors section says Desikan and Killiany developed it. Neither passage establishes that it is the most popularly used cortical parcellation. The later clause about other schemes is cited to other sources, not Fischl (2012). (LLM, conf=0.84)

  cites Fischl, 2012 sourceExperimental — accuracy not yet measured
• Supported

  The Destrieux citation is attached to the clause about other cortical parcellation schemes, not to the claim that Desikan–Killiany is the most popular. The Abstract identifies this paper's distinct sulco-gyral parcellation as available in FreeSurfer, and the Discussion reports that FreeSurfer, including this or an earlier atlas, had been licensed 6,700 times. (LLM, conf=0.90)

  cites Destrieux et al., 2010 source

In the paper

“Other quantitative measures can be computed directly from tractography, such as the volume and the length of a fiber tract (Bajada et al., 2019; Behrman-Lay et al., 2015; Catani et al., 2007; Chandio et al., 2020a; Colby et al., 2012a; Grinberg et al., 2018; Rheault et al., 2020a; Song et al., 2015; Yeatman et al., 2012), have also been used.”

What the sources say

• Partly supported

  Section 2.4 (Length mapping; page not supplied) says the authors calculated streamline lengths directly from tractography track files and used them to create cortical tract-length maps. Our copy does not report computing fiber-tract volume as a quantitative measure; its references to volume in Section 2.3 concern image space and subject brain volume. (LLM, conf=0.88)

  cites Bajada et al., 2019 sourceExperimental — accuracy not yet measured
• Partly supported

  Methods §2.5 says the authors computed mean track length from tractography-derived arc lengths and used it as a quantitative measure. It does not report computing fiber-tract volume; the volume mentioned there is intracranial volume, used to normalize length. (LLM, conf=0.88)

  cites Behrman-Lay et al., 2015 sourceExperimental — accuracy not yet measured
• ?Could not verify

  The tractography methods passage (Discussion; page unavailable) describes measuring the number of reconstructed pathways and fractional anisotropy, but not tract volume or length. Searches for tract volume and fiber length found no passage saying either was computed or used as a quantitative measure in this paper. (LLM, conf=0.75)

  cites Catani et al., 2007 sourceExperimental — accuracy not yet measured

In the paper

“Due to these factors, several studies have demonstrated advantages of fiber clustering methods over the cortical-parcellation-based methods, including more consistent parcellation across subjects (Sydnor et al., 2018; Zhang et al., 2017) and a higher reproducibility between test-retest scans (Zhang et al., 2019).”

What the sources say

• Supported

  Sydnor et al. is cited for more consistent parcellation across subjects, not for test–retest reproducibility, which the sentence attributes to Zhang et al. (2019). In the Discussion section, the paper reports that clustering produced the most consistent white matter tracts across subjects among the methods compared. (LLM, conf=0.92)

  cites Sydnor et al., 2018 source
• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites Zhang et al., 2017
• Supported

  The Zhang et al. (2019) citation is attached specifically to the test–retest claim, not the separate claim about consistency across subjects. The Discussion reports significantly higher test–retest reproducibility for fiber clustering than for the cortical-parcellation-based method across all three datasets. (LLM, conf=0.94)

  cites Zhang et al., 2019 source

In the paper

“For example, researchers have used the longitudinal relaxation rate (R1), a measure sensitive to myelin, along the fiber tracts (Boshkovski et al., 2020a; Yeatman et al., 2012).”

What the sources say

• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites Boshkovski et al., 2020a
• Partly supported

  The Methods section describes measuring fractional anisotropy (FA) at locations along fiber tracts, and the Discussion reports tract profiles of FA. The cited paper therefore supports the along-tract approach, but the passages do not establish that it used longitudinal relaxation rate (R1) or describe R1 as sensitive to myelin. (LLM, conf=0.88)

  cites Yeatman et al., 2012 sourceExperimental — accuracy not yet measured

In the paper

“Most often, data are averaged across corresponding points along each streamline in the pathway, such that data can be analyzed (e.g., an along-tract or along-pathway plot) versus streamline arc length or other parameterization (Colby et al., 2012b; Corouge et al., 2006; Yeatman et al., 2012).”

What the sources say

• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites Colby et al., 2012b
• Supported

  Section 3.2.1 (p. 8) defines points at the same arc length along fibers as corresponding points. Section 3.3.1 (p. 10) computes a mean tensor at each corresponding location, and Fig. 4’s description (p. 12) says the resulting diffusion properties are plotted against location along the tract. This supports the cited paper as an example of the described approach. (LLM, conf=0.92)

  cites Corouge et al., 2006 source
• Supported

  The Methods (Automated Fiber Quantification, Fiber Tract Quantification) say that each fiber is resampled to 100 equally spaced nodes. The Methods then describe averaging diffusion properties across fibers at each node to produce a tract profile, supporting the attributed along-tract analysis. (LLM, conf=0.89)

  cites Yeatman et al., 2012 source

In the paper

“This sampling may be done by first generating a binary mask that defines the voxels through which streamlines pass and then sampling from an underlying microstructure image within this mask (Ciccarelli et al., 2003; Heiervang et al., 2006; Kunimatsu et al., 2004; Voineskos et al., 2010). The sampling may alternatively be done by sampling microstructural measures at each point of the streamlines”

What the sources say

• ?Not checked — no full text of the sourcecites Ciccarelli et al., 2003 source
• ?Not checked — no full text of the sourcecites Heiervang et al., 2006 source
• Supported

  Split into 2 atoms — 2 confirmed, 0 partially supported, 0 disputed, 0 misattributed, 0 unverifiable. Expand the claim to see per-atom verdicts.

  cites Kunimatsu et al., 2004 source

In the paper

“Other approaches fit a medial surface representation to the pathway, such that sampled data can be represented in two dimensions and can be analyzed not only along the pathway but also across its cross-section (Chen et al., 2016b; Qiu et al., 2010; Yushkevich et al., 2009).”

What the sources say

• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites Chen et al., 2016b
• Partly supported

  Methods §4.3 defines the tract surface as the boundary of its volume, not a medial surface. Methods §§4.4–4.5 describe a two-dimensional surface and mapping three-dimensional FA data onto it, supporting the dimensionality-reduction part of the claim. The cited paper does not describe analysis along the pathway and across its cross-section in those terms. Page numbers are unavailable in our copy. (LLM, conf=0.88)

  cites Qiu et al., 2010 sourceExperimental — accuracy not yet measured
• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites Yushkevich et al., 2009

In the paper

“Such analyses have been applied to study, e.g., how white matter fiber tracts are affected in individuals with different disease severity (Cha et al., 2015; Chiang et al., 2016; Price et al., 2007), how tracts relate to cognitive and socio-emotional function (Zekelman et al., 2021), and how tracts are developing during human neurodevelopment and aging (Gertheiss et al., 2013; Hasan et al., 2009a;”

What the sources say

• Supported

  The Effects of ADHD section reports reduced fronto-accumbal tract measures in children with ADHD compared with healthy controls. The Correlation with aggression section also reports that tract measures varied with aggression among children with ADHD, supporting this paper as an example of using tractography to examine differences associated with clinical severity. The later claims about cognition and development are attached to other citations, not Cha et al. (LLM, conf=0.88)

  cites Cha et al., 2015 source
• ?Not checked — no full text of the sourcecites Chiang et al., 2016 source
• Partly supported

  Split into 3 atoms — 0 confirmed, 3 partially supported, 0 disputed, 0 misattributed, 0 unverifiable. Expand the claim to see per-atom verdicts.

  cites Price et al., 2007 sourceExperimental — accuracy not yet measured

In the paper

“Over the last two decades, the study of the brain as a network has become increasingly popular in the neuroscience and neuroimaging communities, giving rise to the new fields of connectomics and network neuroscience (Bassett and Sporns, 2017; Fornito et al., 2013; 2015; 2016).”

What the sources say

• Supported

  The Conclusion describes network neuroscience as a new field gaining momentum as networks have become ubiquitous in neurobiological research. The Network mapping and observation section identifies connectomics as the mapping of connections among neurons and brain areas. Together, these passages support the citing sentence’s broad account of growing network-based research and the emergence of these fields. (LLM, conf=0.88)

  cites Bassett and Sporns, 2017 source
• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites Fornito et al., 2013; 2015; 2016

In the paper

“Additional studies have analyzed tracts represented as surfaces (Chen et al., 2016b; Qiu et al., 2010; Yushkevich et al., 2009) or performed data reduction techniques to reduce the dimensionality of the point-wise measurements along tracts (Ceschin et al., 2015; Chamberland et al., 2019; Geeraert et al., 2020).”

What the sources say

• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites Chen et al., 2016b
• Supported

  Qiu et al. is cited for the first part of the sentence: analyzing tracts represented as surfaces. The Introduction says the study introduced surface representations of 23 deep white matter tracts, and the Discussion identifies construction of those representations as its main contribution. The later clause about data-reduction techniques has different citations and is not attributed to Qiu et al. (LLM, conf=0.95)

  cites Qiu et al., 2010 source
• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites Yushkevich et al., 2009

In the paper

“Construction of the connectivity matrix has been described in several reviews (Sotiropoulos et al., 2016; Yeh et al., 2020) and includes two major steps. First, parcellation of the whole-brain tractogram is needed to compute all possible fiber pathways in the brain, i.e., white matter parcels connecting all pairs of gray matter regions (as described in Section 4.2). Then, a metric of “connectivity”

What the sources say

• Partly supported

  In Results, under “Preserving the fibre spreading pattern to the cortex supported by the 7T data” (Fig. 9; page unavailable), the paper describes constructing a connectome matrix by recording tractography-derived path probabilities between pairs of surface vertices. That supports using a connectivity metric for matrix entries, but the paper does not describe the claimed two-step procedure of parcellating a whole-brain tractogram into pathways between gray-matter regions and then assigning edge weights. It is an application of fibre-orientation estimation, not a review of connectivity-matrix construction. (LLM, conf=0.86)

  cites Sotiropoulos et al., 2016 sourceExperimental — accuracy not yet measured
• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites Yeh et al., 2020

In the paper

“Another tract-specific analysis strategy is data-driven and uses machine learning techniques to perform tasks such as disease classification and prediction (Deng et al., 2019; O’Dwyer et al., 2012; Payabvash et al., 2019; Zhang et al., 2018a), as well as prediction of behavioral measures and traits (Seguin et al., 2020a; Tian et al., 2021).”

What the sources say

• ?Not checked — no full text of the sourcecites Deng et al., 2019 source
• Partly supported

  The Methods and Discussion describe using diffusion data and a support vector machine to classify healthy young APOE4 carriers and non-carriers, and the Introduction says selected voxels were used for tractography. The Discussion also reports related work classifying older subjects as having mild cognitive impairment or being healthy. This paper itself did not classify disease or predict its onset; the later citations in the citing sentence, not O’Dwyer et al., are attached to behavioral measures and traits. (LLM, conf=0.85)

  cites O’Dwyer et al., 2012 sourceExperimental — accuracy not yet measured
• Supported

  Payabvash et al. is cited for disease classification and prediction, not for the separately cited prediction of behavioral traits. The Methods describe supervised machine-learning models using tract-based measures to classify children with sensory processing disorder, and the Results report their classification performance. (LLM, conf=0.93)

  cites Payabvash et al., 2019 source

In the paper

“The disciplines of graph theory and network science provide a framework to quantify network properties of brain connectivity (Rubinov and Sporns, 2010) and investigate whole-brain models of structural wiring (Betzel et al., 2016).”

What the sources say

• ?Not checked — no full text of the sourcecites Rubinov and Sporns, 2010 source
• Supported

  The citation to Betzel et al. is attached to investigating whole-brain models of structural wiring. In the Introduction, the paper describes fitting and comparing generative models of human connectomes to identify wiring rules that reproduce observed networks. (LLM, conf=0.93)

  cites Betzel et al., 2016 source

In the paper

“Selection of tractography filtering strategies is important and has been shown to affect the analysis of the topology of brain networks (Civier et al., 2019; Frigo et al., 2020; Yeh et al., 2016a) and affect sensitivity and specificity when analyzing the connectomes (Zalesky et al., 2016a).”

What the sources say

• Partly supported

  The Introduction says methods used to construct graphs from tractograms can strongly influence connectome characteristics, and §3.2 reports that heavy pruning changes graph-theoretical metrics. But the paper tests removal of connectome connections, not a comparison of tractography filtering strategies; §4.8 says it cannot rule out effects of its fixed pipeline on topology. The sensitivity-and-specificity clause is attributed to Zalesky et al., not Civier et al. (LLM, conf=0.82)

  cites Civier et al., 2019 sourceExperimental — accuracy not yet measured
• Supported

  The Abstract (Main Result) states that tractography filtering techniques change brain-network topology and alter network metrics. The Discussion and Conclusions likewise reports topological changes with SIFT2 and COMMIT. Frigo et al. is cited for this topology claim; the sentence cites Zalesky et al. separately for sensitivity and specificity. (LLM, conf=0.94)

  cites Frigo et al., 2020 source
• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites Yeh et al., 2016a

In the paper

“This modular structure constrains the flow of information through the network in a manner thought to promote functional segregation and clusters of specialized information processing (Betzel et al., 2013).”

What the sources say

• Partly supported

  The Introduction describes communities as preferentially interconnected nodes that can trap the flow of random walkers, a model the paper says can represent information diffusion. The Discussion links connection topology to functional coupling, but our copy does not say that modular structure promotes specialized information processing or functional segregation in those terms. No page numbers were provided for these passages. (LLM, conf=0.82)

  cites Betzel et al., 2013 sourceExperimental — accuracy not yet measured

In the paper

“Popular methods for community detection in structural brain networks include the Louvain algorithm and stochastic block models, though determining best practices around the utilization of these and other approaches remain an active topic of research (Fortunato and Hric, 2016).”

What the sources say

• Partly supported

  The Methods section (p. 22) describes popular community-detection algorithms, and the paper discusses both the Louvain method (p. 30) and stochastic block models (p. 37). The Introduction (p. 2) also says there are no clear-cut guidelines for comparing methods. These passages do not establish that Louvain and stochastic block models are popular specifically for structural brain networks. (LLM, conf=0.86)

  cites Fortunato and Hric, 2016 sourceExperimental — accuracy not yet measured

In the paper

“Graph measures can be classified based on their resolution or scope—from local measures that quantify properties of individual ROIs (nodes), to mesoscale measures describing clusters of interconnected ROIs, to global measures that describe whole-brain connectivity properties (Betzel and Bassett, 2017; Fornito et al., 2016) such as network communication in the brain (Graham et al., 2020) (Fig. 4b).”

What the sources say

• Supported

  Section 3, “Multi-scale network analysis,” distinguishes local measures of individual nodes, global measures of the network as a whole, and an intermediate mesoscale characterized by clusters of nodes. This supports the cited classification by scope; the citing sentence’s ROIs are its application of the paper’s nodes. (LLM, conf=0.94)

  cites Betzel and Bassett, 2017 source
• ?Not checked — no full text of the sourcecites Fornito et al., 2016 source
• Supported

  Graham et al. is cited for the example ‘network communication in the brain,’ not for the preceding classification of graph measures, which is cited to other works. The editorial’s Abstract explicitly discusses communication across whole-brain networks, so it supports that example. (LLM, conf=0.93)

  cites Graham et al., 2020 source

In the paper

“In light of these shortcomings, recent work has stated to focus on alternative models of connectome communication, such as network diffusion (Estrada and Hatano, 2008; Goñi et al., 2014), linear transmission models (Mišić et al., 2015), and navigation (Seguin et al., 2018) (see Avena-Koenigsberger et al., 2018 for a review).”

What the sources say

• Partly supported

  The section “Communicability as the Green’s function of networks” (p. 4) explicitly interprets the model as particle diffusion on a complex network, supporting its citation as an alternative to shortest-path communication. The paper does not study connectomes specifically, so it does not support that setting-specific part of the sentence. (LLM, conf=0.83)

  cites Estrada and Hatano, 2008 sourceExperimental — accuracy not yet measured
• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites Goñi et al., 2014
• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites Mišić et al., 2015

In the paper

“In addition, connectome networks are known to strike a balance between wiring minimization (shortest paths) and formation of connections between brain regions with similar connectivity profiles (homophily) (Betzel et al., 2016).”

What the sources say

• Partly supported

  The Results section reports that the best-fitting model combines a penalty against long-distance connections with a greater probability of connecting regions that have similar connectivity profiles. However, the parenthetical “shortest paths” is inaccurate: the paper describes minimizing physical wiring distance, and its Results > Geometric model section says forming only the shortest connections increases characteristic path length. (LLM, conf=0.88)

  cites Betzel et al., 2016 sourceExperimental — accuracy not yet measured

In the paper

“For instance, despite the inherent limitations of dMRI tractography in resolving axonal fiber directionality, graph measures computed on the undirected human connectome can reveal patterns of asymmetric communication, as evidenced by significant associations to the directionality of effective connectivity computed using dynamic causal modeling (Seguin et al., 2019).”

What the sources say

• Supported

  The Introduction states that communication measures applied to undirected brain networks reveal send-receive asymmetries in the human cortex, and that these asymmetries are significantly associated with effective-connectivity directionality derived from dynamic causal modeling. The Results section, “Senders, receivers, and effective connectivity,” reports significant correlations across pairs of subsystems, with the stated exception of the seven-subsystem partition. (LLM, conf=0.95)

  cites Seguin et al., 2019 source

In the paper

“However, algorithmically, the identification of shortest paths demands individual regions to possess global knowledge of structural connectivity, a requirement unlikely to be met in decentralized systems such as the brain (Avena-Koenigsberger et al., 2019; Goñi et al., 2014; Seguin et al., 2018).”

What the sources say

• Partly supported

  The Introduction says that routing along shortest paths relies on all elements having information about the network’s global topology. The Discussion, however, calls it an open question how a brain system gains that information and raises the possibility that it learns about its topology through feedback; it does not establish that the requirement is unlikely to be met. (LLM, conf=0.86)

  cites Avena-Koenigsberger et al., 2019 sourceExperimental — accuracy not yet measured
• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites Goñi et al., 2014
• Supported

  The introductory passage (listed under Abstract; page not provided) says that identifying shortest paths requires individual elements of nervous systems to possess global knowledge of network topology, and that this requirement is challenged because nervous systems are decentralized. The Discussion section, “Biological Plausibility and Communication Efficiency” (page not provided), likewise describes the centralized knowledge needed to compute optimal routes as biologically implausible. (LLM, conf=0.94)

  cites Seguin et al., 2018 source

In the paper

“Similar studies also show that the radial and tangential pathways organization are prominent at 17 weeks but gradually disappear at later gestational ages, and all major structural pathways are already identifiable by term (Takahashi et al., 2002).”

What the sources say

• ?Could not verify

  The Results describe diffusion imaging of excised larval sea lamprey spinal cords, and the Discussion describes their axonal anatomy. Neither addresses gestational age or whether major structural pathways are identifiable by term. Searches for 17 weeks, radial and tangential pathways, and fetal development found no supporting passage. (LLM, conf=0.78)

  cites Takahashi et al., 2002 sourceExperimental — accuracy not yet measured

In the paper

“Similar studies also show that the radial and tangential pathways organization are prominent at 17 weeks but gradually disappear at later gestational ages, and all major structural pathways are already identifiable by term (Takahashi et al., 2002).”

What the sources say

• ?Could not verify

  The Results and Discussion describe diffusion measurements in excised larval sea lamprey spinal cord; no page numbers are available in our copy. Searches for radial and tangential pathways, 17 gestational weeks, and pathways identifiable by term found no passage addressing fetal development. The lamprey findings therefore cannot verify the attributed claim. (LLM, conf=0.76)

  cites Takahashi et al., 2002 sourceExperimental — accuracy not yet measured

In the paper

“For instance, using ex vivo dMRI scans on fetal brain samples, researchers are able to identify the major tract parts of the fornix and cingulum bundles at 13 weeks, but a part of the corpus callosum at 15 weeks (Huang et al., 2009).”

What the sources say

• Partly supported

  The Results (White matter tracts) say the fornix is appreciable as a major tract at 13 weeks and the corpus callosum appears at 15 weeks. But they say the cingulum bundle can be traced only after 17 weeks, not at 13 weeks. The supplied passages do not give page numbers. (LLM, conf=0.93)

  cites Huang et al., 2009 sourceExperimental — accuracy not yet measured

In the paper

“Due to recent advances in methodological progress such as motion correction algorithms, in vivo tractography on the fetal brain has become available and exhibit highly agreement with ex vivo and neuroanatomical studies (Christiaens et al., 2021; Khan et al., 2019), which indicates strong potential for future prenatal brain investigations.”

What the sources say

• Partly supported

  The Discussion reports that the motion-correction framework has proved effective in fetal data, supporting the methodological-progress part of the attribution. But this paper analyzes scans acquired after birth (Data and preprocessing) and does not report in vivo fetal tractography or compare it with ex vivo or neuroanatomical studies. (LLM, conf=0.85)

  cites Christiaens et al., 2021 sourceExperimental — accuracy not yet measured
• Partly supported

  The Introduction (Fetal DTI; page unavailable) reports that motion-corrected in utero DTI enabled fetal tractography. The Conclusion (page unavailable) says the authors’ atlas portrayed developmental trends seen in prior ex vivo and in utero studies, but it does not establish high agreement specifically between in vivo tractography and ex vivo or neuroanatomical studies. (LLM, conf=0.83)

  cites Khan et al., 2019 sourceExperimental — accuracy not yet measured

In the paper

“revealing a typical spatial degeneration pattern across brain systems in which prefrontal association tracts are the most vulnerable to aging, while cingulum, temporal and parietal-occipital commissural connections are relatively preserved.”

What the sources say

• Partly supported

  The Discussion (Anterior-Posterior Gradient; page unavailable) ranks age-related FA differences highest in the genu and uncinate fasciculus and lowest in the cingulum bundle and splenium, supporting greater vulnerability of anterior tracts and relative preservation of some connections. But the same section reports appreciable effects in the inferior longitudinal fasciculus, including a stronger effect in its occipito-temporal segment; the Results (Anterior-posterior trend analysis; page unavailable) also find age effects in the cingulum. The claim overstates preservation across the named connections. (LLM, conf=0.88)

  cites Davis et al., 2009 sourceExperimental — accuracy not yet measured
• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites Stadlbauer et al., 2008a
• Partly supported

  The Results (section 3.3) identify the frontal forceps as having the most prominent age-related FA decrease, and the Discussion (section 4) describes greater susceptibility in anterior than posterior bundles. But the proposed preservation is too broad: section 4 reports age-related changes in the superior cingulum, while section 3.4 reports significant age effects in temporal and parietal callosal sectors. Page numbers are unavailable in our copy. (LLM, conf=0.87)

  cites Sullivan et al., 2010 sourceExperimental — accuracy not yet measured

In the paper

“Many studies have discovered largely adult-like topological structures including significant small-world, modular, and rich-club organization during the middle to final trimester of gestation”

What the sources say

• ?Not checked — no full text of the sourcecites Brown et al., 2014 source
• Partly supported

  The Results report small-world organization in fetal structural networks at 20, 35, and 40 postmenstrual weeks (Figures 6–7; page not provided). The passages do not establish modular or rich-club organization, or characterize the topology as broadly adult-like. (LLM, conf=0.88)

  cites Song et al., 2017 sourceExperimental — accuracy not yet measured
• Partly supported

  The Results (Changes in Global Topological Properties with Age) report small-world properties at all studied ages and assess modularity. But the study examined premature and term-born neonates, infants, and adults—not fetuses during gestation—and the cited passages do not establish rich-club organization or statistical significance for every claimed property. (LLM, conf=0.86)

  cites Tymofiyeva et al., 2013 sourceExperimental — accuracy not yet measured

In the paper

“tractography studies show that the basic layouts of the brain’s structural network are largely preserved until old age, such as the economical small-world character, the modular organization and rich club organization”

What the sources say

• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites Gong et al., 2009b
• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites Li et al., 2020c
• Partly supported

  The Small-world properties section reports that all studied subjects’ structural networks had small-world architecture, and the Discussion says this property was preserved despite individual variation. The paper does not establish preservation across the lifespan until old age, and the passages found do not report preserved modular or rich-club organization. (LLM, conf=0.88)

  cites Wen et al., 2011 sourceExperimental — accuracy not yet measured

In the paper

“although most tracts show increased FA during childhood and adolescence and decreased FA at older ages”

What the sources say

• Supported

  The Results, under “Age effects” (page not supplied), report an inverted U-shaped FA trajectory for all measured pathways in both males and females. This supports increasing FA at younger ages followed by decreasing FA at older ages within the study’s 6–68-year age range. (LLM, conf=0.93)

  cites Hasan et al., 2010 source
• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites 2009b
• ?Not checked — no full text of the sourcecites Lebel et al., 2010 source

In the paper

“After birth until adulthood, the brain network reshapes to improve integration capacity with the reconfigurations of decreased modularity and decreased small-worldness”

What the sources say

• Partly supported

  The Results (Global network properties and their age-related trajectories; page unavailable) report increased global efficiency and decreased modularity from early to late childhood, with both measures then stabilizing. The Results (Small-world efficiency of developing WM networks; page unavailable) show small-worldness values that fluctuate across age groups rather than a sustained decrease. The study begins at age 6, so it does not establish a trajectory from birth. (LLM, conf=0.88)

  cites Chen et al., 2013 sourceExperimental — accuracy not yet measured
• Partly supported

  The Discussion reports decreasing small-worldness and modularity with age and says the network changes suggest increased structural integration. But the study examined people aged 12–30, so it does not establish that this pattern occurs throughout the period from birth to adulthood. (LLM, conf=0.87)

  cites Dennis et al., 2013 sourceExperimental — accuracy not yet measured
• Partly supported

  The Results report increasing network efficiency and a generally decreasing small-world index with age. But the Modularity and Module Composition results say modularity showed no consistent significant age trend, rather than a decrease. The study examined ages 2–18, not the full period from birth to adulthood. (LLM, conf=0.91)

  cites Hagmann et al., 2010 sourceExperimental — accuracy not yet measured

In the paper

“For instance, in Parkinson’s disease, researchers found that DBS on sub-thalamic nucleus linking to the supplementary motor area is associated with improvement in bradykinesia and rigidity, while DBS on nucleus structurally commenting to M1 is associated with improvement in tremor (Akram et al., 2017).”

What the sources say

• Supported

  Split into 2 atoms — 2 confirmed, 0 partially supported, 0 disputed, 0 misattributed, 0 unverifiable. Expand the claim to see per-atom verdicts.

  cites Akram et al., 2017 source

In the paper

“Therefore, tractography in native space becomes suitable for MS patients because it is free of registration and allows inter-subject variability of tract location and shape (Lipp et al., 2020).”

What the sources say

• Supported

  The Methods (Data analysis) state that tractography was performed in native space for MS patients. The Introduction contrasts individual tractography with probability-map methods that require inter-subject registration and assume consistent tract anatomy; the Discussion (Anatomical generalisability) says tractography is robust to individual differences in tract size and shape. Thus, “free of registration” is supported in the sense of not requiring inter-subject registration for native-space tractography, not as a claim that the study performed no registration for other analyses. (LLM, conf=0.87)

  cites Lipp et al., 2020 source

In the paper

“Furthermore, connectomic neuromodulation may contribute to identifying specific networks that lead to symptom-specific treatments, paving the way for personalized therapy (Horn and Fox, 2020).”

What the sources say

• Supported

  Opportunity #3 (“symptom specific networks”) explicitly says connectomic neuromodulation may identify networks that lead to symptom-specific treatments and that stimulating different networks for different symptoms could pave the way for personalized therapy. The citing sentence preserves the paper’s tentative wording. (LLM, conf=0.95)

  cites Horn and Fox, 2020 source

In the paper

“The reconstructed streamlines are only simulated entities that do not correspond directly to nerve fibers (Jeurissen et al., 2019; Yeh et al., 2020)”

What the sources say

• Supported

  Section 2.3, “The virtual nature of fiber tracking,” explicitly distinguishes physical nerve fibers from reconstructed streamlines, calling the latter virtual entities related only indirectly to nerve fibers. This supports the citing sentence’s paraphrase. (LLM, conf=0.95)

  cites Jeurissen et al., 2019 source
• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites Yeh et al., 2020

In the paper

“clinical studies frequently employ advanced tractography methods to reconstruct individual tract streamlines (Sbardella et al., 2013) and obtain feasible results even when brain lesions are present (Lipp et al., 2020).”

What the sources say

• Supported

  The Introduction (page not provided) describes three-dimensional tractography as reconstructing white-matter pathways by tracking fibers and discusses more sophisticated tracking approaches. This supports the tract-reconstruction point attributed to Sbardella et al.; the separate claim about feasible results in the presence of lesions is attributed to Lipp et al., not to this source. (LLM, conf=0.88)

  cites Sbardella et al., 2013 source
• Supported

  The Lipp citation supports the clause about obtaining feasible results in the presence of lesions. In the Discussion and Conclusions, the paper reports successful individual tract segmentation in a large cohort with MS lesions and says this is feasible even in large-scale clinical studies. The preceding claim about how frequently clinical studies use advanced methods is attached to a different citation. (LLM, conf=0.91)

  cites Lipp et al., 2020 source

In the paper

“basic diffusion metrics are only inferences based on local diffusion properties, which are not direct measures of tissue properties (Assaf et al., 2019).”

What the sources say

• Supported

  In “Local measures of white matter microstructure,” the paper says basic diffusion parameters reflect local microstructure but have no one-to-one relationship with underlying tissue structure. It also says common DTI indices are nonspecific to any particular tissue compartment, supporting the distinction between diffusion-based inference and direct tissue measurement. (LLM, conf=0.92)

  cites Assaf et al., 2019 source

In the paper

“The anatomical accuracy of tractography is an ongoing issue in quantitative tractography analysis, where both false positive and false negative tractography results pose challenges (Maier-Hein et al., 2017; Thomas et al., 2014).”

What the sources say

• Supported

  The Results sections report many invalid connections and bundles (false positives), while no submission recovered all 25 ground-truth bundles and average bundle-volume overlap was only 36% (missed connections or portions of bundles). The Discussion describes tractography’s ambiguity as an open challenge. Page numbers are not available in our copy. (LLM, conf=0.91)

  cites Maier-Hein et al., 2017 source
• Supported

  The Discussion reports that tractography methods with high sensitivity tend to have low specificity and a high rate of false positives. It also explicitly describes missed tracer-labeled regions as false negatives and concludes that anatomical accuracy remains suboptimal even under ideal experimental conditions. (LLM, conf=0.93)

  cites Thomas et al., 2014 source

In the paper

“the fact it can be trivially demonstrated that both that streamline count can be erroneously modulated in the absence of a difference in connection strength and erroneously constant in the presence of a difference in connection strength (Figs. 2, 3)”

In the paper

“Even operating on high-resolution dMRI data, current tractography algorithms have been shown to deviate from the ground-truth fiber projections derived from histological staining, with greater streamline densities at gyral crowns than sulcal banks (Schilling et al., 2018b).”

What the sources say

• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites Schilling et al., 2018b

In the paper

“these “weights” represent an effective cross-sectional area of each streamline, which can be utilized as a direct measure of connection density (Smith et al., 2020a)”

What the sources say

• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites Smith et al., 2020a

In the paper

“This strategy implicitly assumes a correlation between the strength of a reconstructed edge and its validity, which can be violated in a range of ways due to the errors produced by streamlines tractography not being simply random variance.”

What the sources say

• ?Not checked — no full text of the sourcecites Fornito et al., 2016 source

In the paper

“The third point is that anatomical tract identification results can be highly variable across studies due to a lack of consistent fiber tract definitions, where different methods may give different results for the same target anatomical fiber tracts (Dick and Tremblay, 2012; Rheault et al., 2020a; Schilling et al., 2020b).”

What the sources say

• ?Not checked — no full text of the sourcecites Dick and Tremblay, 2012 source
• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites Rheault et al., 2020a
• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites Schilling et al., 2020b

In the paper

“tract-specific tractography may be more computationally efficient if the number of tracts of interest is small, while whole-brain tractography can be computationally expensive but enables tract identification or parcellation of the whole white matter.”

In the paper

“Most of the direct segmentation approaches use a training tract segmentation dataset and thus may not require image registration or require a small amount of work for registration (e.g., a rigid volumetric registration); however, we note that curation of a training tract segmentation dataset may require extensive work compared to performing image registration.”

In the paper

“however, no consensus has been reached given that many existing tract identification methods have not been compared.”

In the paper

“tract-specific tractography may be more computationally efficient if the number of tracts of interest is small, while whole-brain tractography can be computationally expensive but enables tract identification or parcellation of the whole white matter.”

In the paper

“but with more modern multi-fiber tractography, reproducibility is lower, potentially due to the increase in spurious streamlines (Rheault et al., 2020a).”

What the sources say

• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites Rheault et al., 2020a

In the paper

“While using streamline-based registration can avoid the intermodality registration issue in ROI-based methods, streamline registration (especially a nonlinear registration) may be computationally heavier than volumetric registration.”

In the paper

“They parcellate tractography according to a cortical (and sometimes a subcortical) gray matter parcellation, focusing on the structural connectivity among different gray matter ROIs (Bassett and Bullmore, 2017; Bullmore and Sporns, 2009a; Gong et al., 2009a; Ingalhalikar et al., 2014; Sporns et al., 2005; Yeh et al., 2016b; Zalesky et al., 2012).”

What the sources say

• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites Bassett and Bullmore, 2017
• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites Bullmore and Sporns, 2009a
• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites Gong et al., 2009a

In the paper

“Though relatively less used, fiber clustering tractography parcellation methods are increasingly applied to study the brain’s structural connectivity in applications such as disease classification and between-population statistical analysis (Feng et al., 2020; Ji et al., 2019; Wu et al., 2018; Zhang et al., 2018a; 2018b).”

What the sources say

• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites Feng et al., 2020
• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites Ji et al., 2019
• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites Wu et al., 2018

In the paper

“(Note that a whole-brain tractogram is required for filtering (Sections 3.3 and 5.3) and analyses of the brain as a network using a connectivity matrix (Sections 5.5 and 5.6)).”

In the paper

“In this category of measures, tractography is generally used to define the locations in which microstructural measures are sampled.”

In the paper

“There are many quantitative measures that can be computed from tractography, which can be subdivided into two categories based on the source of data used to compute the measures.”

In the paper

“Specifically, tractography segmentation is performed by extracting streamlines that connect pairs of ROIs. Therefore the resulting tractography segmentation is mainly determined by the selection of a cortical parcellation scheme.”

In the paper

“The second category of quantitative measures computed from tractography leverages microstructural measures that are usually computed from a diffusion model (e.g., diffusion tensor model) or another quantitative imaging modality (e.g., myelin imaging).”

In the paper

“Popularly used diffusion-modeling-based microstructural measures include the fractional anisotropy, axial, radial and mean diffusivities (FA, AD, RD and MD respectively) derived from traditional diffusion tensor model (Basser and Pierpaoli, 1996).”

What the sources say

• ?Not checked — no full text of the sourcecites Basser and Pierpaoli, 1996 source

In the paper

“Another hypothesis-driven analysis uses a regression model (e.g., a generalized linear model (Dobson and Barnett, 2018) or support vector regression (Zhang and O’Donnell, 2020)) to assess correlation of the tract quantitative measurement and a behavioral or a disease symptom score.”

What the sources say

• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites Dobson and Barnett, 2018
• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites Zhang and O’Donnell, 2020

In the paper

“Two main strategies can be employed: using a scalar value as a summary statistic, or using data along the length of the fiber pathway.”

In the paper

“The second strategy to quantify an individual fiber pathway is to measure the distribution of the microstructural measures along the fiber pathway.”

In the paper

“In particular, because of their flexible implementations, COMMIT, COMMIT2 and COMMIT2tree allow complementing tractography with biophysical models of the tissue microstructure and, thus, enable one to access more quantitative and biologically informative features of individual bundles, such as average axon diameter (Barakovic et al., 2018), myelin content (Schiavi et al., 2019) and bundle-specific”

What the sources say

• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites Barakovic et al., 2018
• ?Not checked — no full text of the sourcecites Schiavi et al., 2019
• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites Barakovic et al., 2021b

In the paper

“Other summary statistics such as the median, maximum and minimum have also been employed (Boshkovski et al., 2020b; Zhang et al., 2018a; 2018b).”

What the sources say

• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites Boshkovski et al., 2020b
• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites Zhang et al., 2018a
• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites Zhang et al., 2018b

In the paper

“For example, compared to the mean statistic, studies have shown that the median can be more robust against outliers and does not rely on the normality assumption for the distribution of the microstructure parameter along a streamline (Boshkovski et al., 2020b; Zhang et al., 2018b).”

What the sources say

• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites Boshkovski et al., 2020b
• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites Zhang et al., 2018b

In the paper

“These methods have been used to investigate properties of healthy and pathological brains, showing better performance than the standard NOS (Schiavi et al., 2020b; Smith et al., 2015a; Yeatman et al., 2014b).”

What the sources say

• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites Schiavi et al., 2020b
• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites Smith et al., 2015a
• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites Yeatman et al., 2014b

In the paper

“While in its original implementation SIFT (Smith et al., 2013) provides only the NOS adjusted based on voxel-wise spherical deconvolution, its evolution SIFT2 (Smith et al., 2015b) optimizes per-streamline cross-section multipliers to match a whole-brain tractogram to fixel-wise fiber densities and thus provides a value for each streamline that can be used (for example) to define the edge weights”

What the sources say

• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites Smith et al., 2015b

In the paper

“The choice of the best summary statistic for data measured within a fiber pathway is still open.”

In the paper

“These methods may rely on a summary statistic per tract, or analyze data along a tract.”

In the paper

“Once a quantitative measure has been extracted for individual fiber pathways, several options exist for tract-specific analysis, which may be hypothesis-driven using statistical analysis or data-driven using machine learning.”

In the paper

“One study has suggested that the maxima and minima are more discriminative than the mean value in machine-learning-based disease classification (Zhang et al., 2018a).”

What the sources say

• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites Zhang et al., 2018a

In the paper

“The underling assumption comes from the fact that a streamline reconstructed from tractography cannot represent a single axon, but rather a group of axons following the same trajectory, and thus we can suppose that on average the microstructure properties the magnetic resonance signal is sensitive to at achievable resolution remain constant.”

In the paper

“(This is required for connectome-based analysis and is a popular approach for tract-specific analysis).”

In the paper

“One important point to note is that correction for multiple comparisons is needed if there are multiple tracts and/or multiple quantitative measurements analyzed.”

In the paper

“Other approaches can employ streamlines of the entire tract in the statistical analyses (Chandio et al., 2020b).”

What the sources say

• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites Chandio et al., 2020b

In the paper

“Unlike statistical analysis methods that usually aim to find white matter structures with statistical group differences, the machine-learning-based methods aim to offer predictive relevance.”

In the paper

“In machine learning analysis, the quantitative measures computed from individual fiber pathways are treated as feature descriptors, which are input to a machine learning algorithm (e.g., a support vector machine) to train a model from a set of training samples with known information (e.g., labels such as disease or healthy control).”

In the paper

“This step is typically done by simply thresholding according to some rule or to reach a predefined density, or using filtering techniques described in Section 5.3.”

In the paper

“Selection of a gray matter parcellation scheme (as described in Section 4.2) is the first essential consideration when constructing a structural connectivity matrix, and it has been shown to be highly consequential in connectome analysis (de Reus and Van den Heuvel, 2013; Yeh et al., 2020; Zalesky et al., 2012).”

What the sources say

• ?Not checked — no full text of the sourcecites de Reus and Van den Heuvel, 2013 source
• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites Yeh et al., 2020
• ?Not checked — no full text of the sourcecites Zalesky et al., 2012 source

In the paper

“Along-tract analysis enables the study of local regions along the tract.”

In the paper

“For scalar values that are interpreted as inversely related to connectivity “strength,” such as RD, MD, or maps reflecting extracellular or isotropic/free water components of the signal, a transformation of the connectivity matrix may be performed using the reciprocal or the log function prior to any subsequent connectivity analysis (Fornito et al., 2016).”

What the sources say

• ?Not checked — no full text of the sourcecites Fornito et al., 2016 source

In the paper

“Another advantage of machine learning is that it can be easily applied to multimodal data.”