New insights into the pathology of white matter tracts in cerebral palsy from diffusion magnetic resonance imaging: a systematic review


Associate Professor Stephen Rose, University of Queensland Centre for Clinical Research, Royal Brisbane and Women’s Hospital, Brisbane 4029, QLD, Australia. E-mail


Aim  Structural connectivity analysis using diffusion magnetic resonance imaging (dMRI) and tractography has become the method of choice for studying white matter pathology and reorganization in children with congenital hemiplegia. To evaluate its role in the research domain, we systematically reviewed the literature about children with cerebral palsy (CP) to document common findings and identify strengths and possible limitations of this neuroimaging technology.

Method  A literature search was performed for peer-reviewed studies pertaining to dMRI and CP.

Results  Twenty-two studies met the inclusion criteria. The corticospinal tract was studied in greatest detail (18/22). The most common finding was decreased fractional anisotropy and/or increased mean diffusivity, indicating significant loss in the integrity of these corticomotor pathways. Fewer studies assessed ascending sensorimotor pathways including the posterior and superior thalamic radiations, which also showed decreased fractional anisotropy. Anisotropy indices (fractional anisotropy, mean diffusivity) obtained for both corticomotor and sensorimotor tracts were repeatedly shown to correlate with clinical measures. Other tracts studied included commissural and association fibres, which showed conflicting results.

Interpretation  There is sound evidence that dMRI-based connectivity techniques are useful for improving our understanding of the structure–function relationships of corticomotor and sensorimotor neural networks in CP.


Corticospinal tract


Posterior limb of the internal capsule


Posterior thalamic radiations


Superior thalamic radiations


Region of interest

What this paper adds

  •  Summary of current evidence for corticomotor and sensorimotor tract integrity in CP including clinical correlations.
  •  Identification of the need to investigate neural networks associated with cognitive domains because of limited current literature or conflicting findings so far.
  •  Identification of discrepancies and limitations in current studies and recommendations for future research.

Cerebral palsy (CP) is a heterogeneous group of non-progressive brain pathologies manifesting as motor, sensory, cognitive, and communication difficulties.1 It affects an estimated 1 to 2.5 per 1000 live births in Western countries.2 There is no cure for CP; current research is aimed at prevention, early detection, and rehabilitation. The pathological features of brain development in this group of disorders have been repeatedly shown to correlate with functional and clinical outcomes for the affected children.3–6 Although the clinical features of CP allow some insight into the timing and aetiology of the insult, neuroimaging can provide additional information about the location and extent of injury.7 Multiple neuroimaging modalities are used both clinically and for research to probe the structural integrity of various parts of the brain in vivo.8 Conventional structural magnetic resonance imaging (sMRI) is the most widely used tool for assessing gross brain pathology in CP. Brain pathology can be described as being normal, maldeveloped, having predominately grey matter lesions, periventricular white matter lesions, or not fitting any of these descriptions, as described by Krageloh-Mann and Horber.4 There are limitations of the use of sMRI for diagnosis and classification of children with CP. Systematic reviews have shown 14 to 17% of children with functional impairment have no abnormality on T1- and T2-weighted MRI.4,9 This is largely due to the inability of sMRI to probe the microstructural integrity of the underlying brain tissue. There is consequently sound motivation to investigate more advanced, non-invasive imaging techniques to improve our understanding of the relationship between brain structure and function in children with CP.

Diffusion MRI (dMRI) uses imaging sequences that are sensitive to the motion of water molecules.10 In this approach, diffusion anisotropy is measured along several orientations with earlier studies, typically using six directions.11 The introduction of improved imaging hardware and image analysis technology has enabled more accurate measures of diffusion, with some studies using up to 64 diffusion-encoding directions.12 Although the resolution of dMRI is dependent on the acquisition sequence and scanner performance, the typical resolution of image voxels ranges from sizes of 2 to 2.5mm3. The most common method used to study diffusion processes in the brain is diffusion tensor imaging. In this case, the diffusion of water is mathematically modelled using a tensor approach, whereby the shape of the resulting diffusion ellipsoid is described by the resulting primary eigenvectors and eigenvalues.13 From these indices, quantitative measures of diffusion anisotropy, such as the fractional anisotropy, can be calculated.14 This summary measure, along with other quantitative diffusion indices such as the directionally averaged mean diffusivity (also referred to as average apparent diffusion coefficient, abbreviated as MD, Dav or ADCave) have proved useful for probing the integrity and development of white matter pathways in the brain.15 The value of fractional anisotropy ranges between 0 and 1, which gives an indication of how uniform (anisotropic) the directionality of diffusion of water molecules is within the given voxel.14 The mean diffusivity is calculated by taking the mean of measured apparent diffusion coefficients from each direction measured in a voxel.15 It is therefore important to know both of these values as they confer different findings. For example, a tract may have minimal diffusivity, but still have a high anisotropy if the diffusion is unidirectional. Conversely, a tract may have high diffusivity, but a low anisotropy if the diffusion is more isotropic. Specific types of white matter pathology are known to have specific diffusion ‘signatures’: for example Wallerian degeneration shows decreased fractional anisotropy with little change in mean diffusivity.16 These measures are typically reported for one or more particular region of interest (ROI), normally drawn manually on sMRI or fractional anisotropy colour maps using standardized guidelines, or, in more recent studies, using an automated processes involving novel brain atlases.17,18 Diffusion tensor imaging can also be exploited to map the trajectories of white matter fibre tracts in the brain using tractography-based algorithms.19 Using this approach, specific white matter pathways can be independently investigated by tracking the coherent diffusion properties in adjacent voxels using a streamline framework15 (Fig. 1). Currently, this imaging strategy is the only non-invasive method for mapping white matter architecture in the living brain (other methods, such as transcranial magnetic stimulation allow for mapping of cortical areas). Diffusion measures of fractional anisotropy and mean diffusivity can then be assessed within specific white matter pathways.20 Additionally, the number of streamlines generated by the tractography algorithm can be reported as a measure of connectivity within the tract.21 More recent alternatives to the diffusion tensor model have been shown to be reliable and reproducible.22 These approaches, which use optimized acquisition schemes such as high angular resolution diffusion imaging23 and higher-order modelling of diffusion anisotropy, allow improved resolution of crossing fibres within each voxel, thus affording more accurate estimations of tract injury or plasticity within corticomotor networks.12 Several excellent review articles have been published which outline the latest techniques for resolving crossing fibres using dMRI.24–27

Figure 1.

 Anisotropy and tractography maps from a participant with right congenital hemiplegia. Representative images showing (a) fractional anisotropy, (b) mean diffusivity, (c) colour-coded fractional anisotropy (green, tracts projecting in an anterior–posterior orientation; red, left–right; blue, superior–inferior orientation), and (d) corticospinal tracts for a participant with right hemiplegia. The corticospinal tractography map was generated using a fully automated image processing pipeline.12 R, right; L, left.

The use of dMRI and tractography in CP is a novel way to elucidate further insights into its pathogenesis. This allows specific insights into white matter anatomy, and identification of tracts with altered integrity compared with the brains of typically developing children. In the early 1960s, the corticospinal tract (CST) was classically thought to be the primary tract involved with functional motor impairment following autopsy studies.28 This was challenged by Hoon et al.29 in 2002 with the proposal that connectivity in ascending sensory pathways may be more compromised than descending motor pathways. Consequently there have been many studies probing the integrity of several tracts in different types of CP, both ascending and descending, with several studies finding that sensorimotor thalamic pathways may have more influence on sensory and motor function than descending corticomotor pathways.12,30 There has been increasing interest in correlations between tract damage and clinical outcomes, as this information may hold prognostic value, and guide specific early therapy. Clinical correlations investigated typically include one or more of global motor function, including standardized schemes such as the Gross Motor Function Classification System (GMFCS),31 and upper limb motor and sensory function.

The aim of this review is to identify and integrate systematically the evidence provided by cross-sectional cohort studies and case–control studies using diffusion imaging in children with CP, and thereby elucidate which white matter tracts are most likely to be involved with a quantitative clinical deficit. This information may be of use to guide early intervention and therapy targeted towards infants at risk of developing CP, which has been repeatedly shown to be effective, as well as improving quality of life for these children.32


Search terms

A literature search was conducted of relevant databases (PubMed, Embase, Cinahl, Scopus, and PsycINFO) on 29 February 2012 for the keywords ‘cerebral palsy’ and any of ‘tractography’, ‘diffusion imaging’, ‘diffusion magnetic resonance imaging’, ‘diffusion tensor imaging’, ‘high angular resolution diffusion imaging’, or ‘diffusion weighted imaging’ (including associated acronyms). Only peer-reviewed publications in English were considered. A full protocol is available as Appendix SI (supporting information published online).

Inclusion and exclusion criteria

Studies were included if they met the following criteria: (1) the study type was a cross-sectional cohort or case–control study; (2) at least one participant group were clinically diagnosed as having CP; and (3) diffusion MRI (either dMRI, diffusion tensor imaging, or newer techniques) had been performed on all included participants, with or without tractography.

Consequently, studies were excluded if they: (1) did not pertain solely to CP; (2) were studies of acute or traumatic brain injury; or (3) were studies of animal models.

All studies meeting these criteria were included, with any potential bias due to technology, methodology, or participant selection being reported and discussed below. Single case studies were not included as methodologies, equipment, and reported variables in such studies are extremely heterogeneous, and therefore a comparison between groups is only valid within each specific study protocol.

Data extraction

Data were extracted independently, with reference to imaging parameters used (magnet strength, B value, number of directions used), methodology used (diffusion tensor imaging, high angular resolution diffusion imaging, tractography, tract-based spatial statistics, method of ROI placement), participants and controls (subtypes of CP, age, sex, number of participants/controls), clinical measures used, ROIs or white matter tracts included, and finally imaging parameters reported with statistical power between groups. Results were then categorized into descending corticomotor tracts, ascending sensorimotor tracts, and commissural and association tracts. As hemiplegia was studied in significantly more detail than any other subtype of CP, all asymmetry findings in hemiplegia were additionally reported. As results across different scanners, varying acquisition parameters, and heterogeneous analysis methodologies are not directly comparable, a meta-analysis was not possible. The most valuable data from each study are therefore the statistical significance of any differences noted between groups or hemispheres within each study. Statistical p-values have been included where reported. The method for calculation of these values varies within each study according to methodology. For instance, tract-based spatial statistics uses a cluster-based statistical analysis within the analysis software,33 whereas ROI or tract-based analysis is largely study specific, and may require corrections for multiple comparisons depending on variables analysed. As such, comparison of p-values between studies is a significant challenge and stated p-values should be interpreted cautiously.


The initial search on 29 February 2012 returned 162 unique results. Of these, 20 studies met strict inclusion criteria. Two further peer-reviewed articles12,34 that did not occur in the initial search but were known to the authors also met inclusion criteria. Details of the included studies, including limitations, are outlined in Table I.

Table I.   Details of included papers
StudyImagingAnalysisnCP typeAge range (mean)Sex, F:MCtrlsClinical measuresPrinciple findingsLimitations
  1. Imaging, acquisition parameters (strength of magnet; B, B value; dirs, number of directions used); Ctrls, number of age-/sex-matched comparison children; CP, cerebral palsy; F:M, females:males; ?, value not reported; ROI, region of interest; ↓, Decreased; ↑, Increased; GMFCS, Gross Motor Function Classification System; UL, upper limb function; GF, global function; CST, corticospinal tract; PLIC, posterior limb of the internal capsule.

Arzoumanian et al.501.5T
B: 1000
6 dirs
Fractional anisotropy and mean diffusivity analysis of manually drawn ROIs (using fractional anisotropy and mean diffusivity maps)3Hemiplegia (n=?)
Diplegia (n=?)
34–42wks (?)?50None↓ Fractional anisotropy in PLIC of CP groupLimited participant numbers
Six directions
Heterogeneous CP group
No ROI reliability analysis
Bleyenheuft et al.511.5T
B: 800
16 dirs
Symmetry of manually drawn ROIs (using fractional anisotropy colour maps) in axial planes12Hemiplegia10–16y (12y 6mo)2:1012Other (UL)Asymmetry in CST correlates with sensory, motor, and functional abilitiesNo ROI reliability analysis
Chang et al.491.5T
B: 1000
32 dirs
Fractional anisotropy and mean diffusivity analysis of manually drawn ROIs (using fractional anisotropy colour maps) and tracts (using probabilistic tractography)23Diplegia (n=12)
Quadriplegia (n=11)
? (2y 10mo)11:1212GMFCS, other (GF)↓ Fractional anisotropy and ↑ mean diffusivity in CST of quadriplegia group compared with diplegia group. ↓ fractional anisotropy and ↑ mean diffusivity in upper limb specific motor ROIs only in quadriplegia groupNo ROI reliability analysis
Faria et al.441.5T
B: 700
31/34 dirs
Automated parcellation of CP brain anatomy using diffusion properties13Hemiplegia (n=1)
Diplegia (n=11)
Quadriplegia (n=1)
4–13y (6y 5mo)5:835NoneNovel technique for automated parcellation of heterogeneous CP group using diffusion imaging. Demonstration of automated abnormality detection.Novel technique used on relatively small heterogeneous group
Faria et al.451.5T
B: 700
32 dirs
Automated parcellation of brain anatomy using fractional anisotropy. Automated detection of abnormalities using this technique13Not stated3.3–13.9y (7y)5:835None↓ Fractional anisotropy in the CST of the pons for CP groupHeterogeneous CP group
Novel technique
Glenn et al.341.5T
B: 1000
6 dirs
Fractional anisotropy and mean diffusivity analysis of tracts (using deterministic tractography) seeded using manually drawn ROIs (using b0 image). Statistical analysis of asymmetry15Hemiplegia6mo–17y (2y)7:817Other (GF)Clinical severity of hemiplegia correlated with ↓ fractional anisotropy and ↑ mean diffusivity.Six directions
No ROI reliability analysis
Holmström et al.351.5T
B: 1000
45 dirs
Fractional anisotropy and mean diffusivity analysis of manually drawn ROIs (using fractional anisotropy colour maps) and tracts (using probabilistic tractography)15Hemiplegia7y 2mo–17y (12y 4mo)9:624GMFCS, other (UL)↓ Fractional anisotropy in cerebral peduncle, PLIC and tract between the two
Fractional anisotropy of contralateral tract correlated manual dexterity scores
Hoon et al.291.5T
B: 600
6 dirs
Qualitative analysis of fractional anisotropy colour maps and 3D tract reconstructions (using deterministic tractography)2Diplegia6–6y (6y)0:22NoneQualitatively identified posterior thalamic radiations as tract most different between CP group and controls. CST did not appear markedly different from controlsLimited participant numbers
6 directions
Qualitative analysis
Hoon et al.301.5T
B: 700
30 dirs
Semi-qualitative analysis of fractional anisotropy colour maps28Hemiplegia (n=2)
Diplegia (n=21)
Quadriplegia (n=4)
Athetosis (n=1)
16mo–13y (5y 10mo)12:1635Other (UL)More severe injury identified in posterior thalamic radiations than CST. Posterior thalamic radiation injury correlated with clinical motor and sensory scoresHeterogeneous CP group
Qualitative measures (reliability analysis previously published)
Koerte et al.361.5T
B: 1000
6 dirs
Fractional anisotropy and mean diffusivity analysis of manually drawn ROIs (using fractional anisotropy colour maps) and tracts (using deterministic tractography)7Diplegia10–18y (15y 2mo)4:312Other (UL)↓ Fractional anisotropy in the transcallosal motor fibres of CP group, but not CSTSix directions
Lee et al.433T
B: 600
45 dirs
Tract-based spatial statistics used to locate voxels where clinical measures correlated with fractional anisotropy43Diplegia6–29y (13y)17:2643GMFCSGMFCS correlated with ↓ fractional anisotropy in CST more than thalamocortical tractsUse of GMFCS as continuous scale
Murakami et al.371.5T
B: 1000
15 dirs
Fractional anisotropy and mean diffusivity analysis of manually drawn ROIs (using fractional anisotropy colour maps) and tracts (using deterministic tractography)5Hemiplegia (n=1)
Diplegia (n=3)
Quadriplegia (n=1)
9mo–41m (19mo)?5None↓ Fractional anisotropy in CST of CP groupLimited participant numbers
Heterogeneous CP group
No ROI reliability analysis
Nagae et al.461.5T
B: 700
30 dirs
Semi-qualitative analysis of fractional anisotropy colour maps24Hemiplegia (n=2)
Diplegia (n=18)
Quadriplegia (n=3)
Athetosis (n=1)
16mo–13y3m (6y)10:1435NoneMost frequent injury identified in retrolenticular part of internal capsule, posterior thalamic radiations, superior corona radiate and commissural fibresHeterogeneous CP group
Qualitative measures (with reliability analysis)
Rha et al.473T
B: 600
45 dirs
Fractional anisotropy, mean diffusivity and streamline count analysis of tracts (using deterministic tractography) seeded using manually drawn ROIs (using fractional anisotropy colour maps)19Bilateral spasticity8mo–2y 11mo (1y 8mo)9:100GMFCS↓ Streamline count in the CST of children in GMFCS level IV–V compared with I–IIINo comparison group
Rose et al.123T
B: 3000
60 dirs
Asymmetry analysis of streamline count (using probabilistic tractography) seeded using automatically parcellated ROIs16Hemiplegia? (10y 7mo)11:50Other (UL)↓ Streamline count in affected CST. Streamline count of sensory tracts correlated with hand function more than motor tractsNo comparison group
Son et al.381.5T
B: 600
32 dirs
Fractional anisotropy, mean diffusivity and asymmetry analysis of manually drawn ROIs (using fractional anisotropy colour maps) and tracts (using deterministic tractography)4Hemiplegia11mo–7y (2y 6mo)1:34NoneAsymmetry of fractional anisotropy and mean diffusivity in CST only apparent at level of periventricular white matter damageLimited participant numbers
No ROI reliability analysis
Son et al.391.5T
B: 600
32 dirs
Fractional anisotropy, mean diffusivity and streamline count analysis of tracts (using deterministic tractography) seeded with manually drawn ROIs (using fractional anisotropy colour maps)2Hemiplegia3–6y (4y 6mo)1:16Other (GF)↓ fractional anisotropy and ↓ streamline count in affected CST of CP group 2–5y before clinical diagnosis of CPLimited participant numbers
No ROI reliability analysis
Thomas et al.401.5T
B: 900
15 dirs
Fractional anisotropy, mean diffusivity and streamline count analysis of manually drawn ROIs (using fractional anisotropy colour maps) and tracts (using deterministic tractography)5Hemiplegia12–16y (14y)2:15None↓ Streamline count in CST, corticobulbar tract and superior thalamic radiation in contralateral side. ↑ Streamline count in ipsilateral corticobulbar tractLimited participant numbers
No ROI reliability analysis
Trivedi et al.411.5T
B: 1000
10 dirs
Fractional anisotropy and mean diffusivity analysis of tracts (using deterministic tractography) seeded using manually drawn ROIs (using T2 images)39Quadriplegia?9:3014GMFCSGMFCS correlated with both ↓ fractional anisotropy and ↑ mean diffusivity of both motor and sensory tracts 
Trivedi et al.521.5T
B: 1000
10 dirs
Fractional anisotropy and mean diffusivity analysis of manually drawn ROIs (using fractional anisotropy colour maps)8Quadriplegia3–12y (6y 2mo)1:710GMFCS, other (GF)↑ Fractional anisotropy in CST of CP group after botulinum toxin A administration and physiotherapyNo ROI reliability analysis
Yoshida et al.481.5T
B: 500
12 dirs
Fractional anisotropy and mean diffusivity analysis of manually drawn ROIs (using fractional anisotropy colour maps)45Athetosis (n=19)
Spasticity (n=26)
6mo–15y (?)13:3231None↓ Fractional anisotropy and ↑ mean diffusivity more pronounced in athetosis than spasticity in multiple structuresNo ROI reliability analysis
Yoshida et al.421.5T
B: 1000
12 dirs
Fractional anisotropy and mean diffusivity analysis of manually drawn ROIs (on fractional anisotropy colour maps) and tracts (using deterministic tractography)34Hemiplegia (n=8)
Diplegia (n=19)
Quadriplegia (n=6)
Triplegia (n=1)
5mo–8y (2y 2mo)14:2021GMFCS↓ Streamline count and ROI-based fractional anisotropy in CST and posterior thalamic radiations. GMFCS correlated with both motor and sensory parametersHeterogeneous CP group
No ROI reliability analysis
Use of GMFCS as continuous scale

Imaging parameters

Most studies (21) reported findings using the diffusion tensor model. Only one study used high angular resolution diffusion imaging with a higher-order diffusion model.12 Thirteen studies used tractography12,34–42 and one study used tract-based spatial statistics43 to probe the integrity of white matter pathways. Three studies used 3T MRI scanners,12,43 whereas the remaining 19 studies used 1.5T MRI scanners. Included studies used B values ranging from 500 to 3000 s/mm2 (median 850s/mm2; mode 1000s/mm2). Diffusion was measured in several directions ranging from six to 60 (median 23; mode 32). Of the 20 studies that used ROI placement, 17 used manual placement, whereas three used automatic parcellation.12,44,45 Only six of the studies using manual ROI placement included reliability analysis.30,35,36,41,46,47

Participants and clinical measures

All 22 studies included at least one patient group consisting solely of participants with CP. Twenty of these studies compared this group with a typically developing comparison group. Four studies included comparisons between groups: children with athetosis versus spasticity;48 different GMFCS levels;41,47 and children with quadriplegia versus diplegia.49 The number of participants ranged from 2 to 45 (mean 17.1). Ages of participants ranged from 5 months to 29 years (only one study included participants over age 18 [Lee et al.43]) with mean range being 3 years 9 months to 12 years, and overall mean age being 7 years. There were on average 1.6 males to each female. Four studies included participants with spasticity in any motor distribution;37,42,45,50 14 studies investigated specific motor distributions of spasticity, including seven (32%) of children with spastic hemiplegia;12,34,35,38–40,51 two studies of children with spastic diplegia;36,43 three studies of children with spastic quadriplegia;29,41,52 one study of children with bilateral spasticity;47 and one study comparing spastic quadriplegia with diplegia.49 One study compared participants with athetosis and spasticity;48 the remaining three studies included all types of CP30,46 or did not specify which subtypes were included.45 Twelve studies assessed correlations between imaging and clinical outcomes.12,30,34–36,39,41–43,47,51,52 Of these, six assessed GMFCS,35,41–43,47,52 three used other clinical measures of global function,34,39,52 and five assessed upper limb function.12,30,35,36,51

Corticospinal tract

The neural network most frequently investigated was the CST. Results are summarized in Table II. Eighteen studies reported diffusion properties of the CST.12,29,30,34–42,45,47–49,51,52 All studies reported at least one significant difference between CP and comparison groups. Twelve studies used tractography to assess tract-based fractional anisotropy of the CST,34–42,45,47,49 of which 10 reported significantly decreased fractional anisotropy in CP.34,35,37–42,45 Five studies used ROI-based analysis of CST at one or more locations,37,40,42,48,52 of which four reported significantly decreased fractional anisotropy in participants with CP.40,42,48,52 No study reported increased fractional anisotropy by either method. Eight studies used tractography to assess mean diffusivity of the CST,34–36,38,40,41,47,49 of which six reported increased mean diffusivity in the participant group.34,36,38,40,41,49 Three studies used ROI-based analysis to assess mean diffusivity of the CST,40,48,52 of which one showed increased mean diffusivity in CP (p<0.001).52 No study showed decreased mean diffusivity. Three studies assessed the volume or cross-sectional area of the CST,45,46,51 and all reported a decrease in children with CP. Four studies also reported on the number of tracts generated using tractography,12,39,42,47 and all reported a decrease in the number of tracts in CST of CP groups. Positive correlation was shown between fractional anisotropy of the CST and motor function as measured by GMFCS in participants with spastic diplegia (p<0.03)43 and spastic quadriplegia (p<0.001)41 as well as qualitative severity of spastic hemiplegia.34 Additionally, it was shown that children with quadriplegia displayed significantly lower fractional anisotropy and higher mean diffusivity than children with diplegia (p<0.025).49 One study looking at all motor distributions of spasticity in CP showed that the GMFCS correlated well with fractional anisotropy of the CST when assessing ROI-based fractional anisotropy (p<0.001) but not tract-based fractional anisotropy.42 Only one study, which included all types of CP, found no significant correlation between CST diffusion parameters and sensory or motor outcomes.30 This study used a qualitative scheme to assess tract integrity, whereas all conflicting studies used quantitative measures. It was shown in spastic quadriplegia that fractional anisotropy of the CST is increased after treatment with botulinum toxin A and physiotherapy (p<0.001).52

Table II.   Diffusion properties of the corticospinal tract
MethodMeasureTotaln (%)n (%)n (%)
  1. Number of tracts, number of tracts generated using tractography; CP>comparison, significantly higher value in group with cerebral palsy; CP=comparison, no significant difference between comparison and cerebral palsy groups; CP<comparison, significantly lower value in group with cerebral palsy. CP, cerebral palsy.

Region of interestFractional anisotropy 50 (0)1 (20) 4 (80)
Mean diffusivity 31 (33)2 (67) 0 (0)
TractographyFractional anisotropy120 (0)2 (17)10 (83)
Mean diffusivity 85 (75)2 (25) 0 (0)
Number of tracts 30 (0)0 (0) 3 (100)

Other descending corticomotor tracts

Other descending tracts studied included the corticobulbar tract and the posterior limb of the internal capsule (PLIC). Three studies reported diffusion properties of the corticobulbar tract.40,41,46 One of these assessed volume only, reporting a decreased volume in CP.46 The other two studies both reported a reduced fractional anisotropy and increased mean diffusivity in CP,40,41 consistent with findings in the CST. Three studies showed significantly reduced fractional anisotropy in the PLIC using ROI-based analysis.35,42,48 One of these also reported an increased mean diffusivity (p=0.002).35 No study reported any findings to the contrary; however, one study assessed qualitatively that the PLIC was relatively intact compared with other tracts.46 The fractional anisotropy of the PLIC was not different between participants with athetosis and spasticity; however, mean diffusivity was significantly increased only in the group with athetosis.48 Fractional anisotropy in the PLIC was shown to correlate with upper limb dexterity as measured by the Box and Blocks test (p=0.027).35

Ascending sensorimotor tracts

Two studies demonstrated significantly decreased fractional anisotropy in the posterior thalamic radiations (PTR) of participants with CP.42,48 In addition two studies noted the PTR to be among the most injured tracts qualitatively.29,46 One study found no difference in fractional anisotropy or mean diffusivity between controls and children with hemiplegia,40 and another showed no difference in fractional anisotropy, mean diffusivity, or number of tracts between controls and children with diplegia.47 Two studies compared clinical correlations between the PTR and CST.30,42 One study showed PTR injury to be correlated with both sensory and motor outcomes including proprioception (right side only) and ambulation (p=0.02), while finding no correlation with CST injury.30 Another found ROI-based (but not tract-based) fractional anisotropy of both the CST (p<0.001) and PTR (p=0.008) to correlate with the GMFCS; however, CST was found to be the stronger predictor.42 Two studies reported on the superior thalamic radiations, both reporting an increased mean diffusivity.40,41 One study reported no change in fractional anisotropy,40 whereas the other reported decreased fractional anisotropy only in GMFCS level V (highest impairment: ‘transported in a manual wheelchair’; p<0.05).41

Commissural and association tracts

Five studies assessed the diffusion properties of the corpus callosum.36,37,40,46,48 Results were conflicting, with two studies showing no significant difference between controls and CP. (One of these studies40 did show a difference [p=0.008] on ROI-based analysis but not on tract-based analysis; the other37 did not show a significant difference by either method.) Of the remaining studies, one reported significant qualitative damage to the corpus callosum,46 whereas the other two both reported reduced fractional anisotropy and increased mean diffusivity compared with controls.36,48 Participants with athetosis had further reduced fractional anisotropy (p<0.01) and increased mean diffusivity (p<0.01) in the right genu of the corpus callosum compared with spasticity.48 One study looked particularly at transcallosal motor fibres,36 and showed reduced fractional anisotropy (p=0.01) and increased mean diffusivity (p=0.001), more pronounced than in the CST. One study43 showed that fractional anisotropy of the corpus callosum decreased with increasing global impairment, as measured by the GMFCS (p<0.05). Two studies assessed association fibres.46,48 One mentioned no significant qualitative difference between CP and controls in association fibres;46 the other showed significantly reduced fractional anisotropy and increased mean diffusivity in the superior longitudinal fasciculus (p<0.001), significantly more pronounced in children with athetosis than those with spasticity.48 The inferior longitudinal fasciculus was not significantly different from controls, with the exception of reduced fractional anisotropy on the left side only in the children with athetosis (p=0.01).

Spastic hemiplegia

Four studies used diffusion tensor imaging to identify the CST and assess symmetry of properties between ipsilateral and contralateral tracts in spastic hemiplegia.12,34,38,51 Asymmetry was demonstrated in cross-sectional area,51 fractional anisotropy,34,38 mean diffusivity,38 and number of tracts.12 Asymmetry was shown to have significant correlation with upper limb sensory function as measured by stereognosis (p<0.001), dexterity (p=0.009), and manual ability as measured by the ABILHAND-kids rating system (p=0.014)51 as well as overall qualitative severity of hemiplegia.34 One study showed no significant correlation between CST asymmetry and impaired limb function but did show significant correlation with sensorimotor thalamic projections (p=0.006).12 A correlation was shown between fractional anisotropy in the contralateral CST and upper limb dexterity as measured by the Box and Blocks test (p=0.007).35 A compensatory hypertrophy of the ipsilateral tract was not demonstrated;35 however, an increased fractional anisotropy was found in the ipsilateral tract compared with controls.40 One study showed in two participants that decreased fractional anisotropy and fibre count in the CST in early childhood were present before manifestation of clinical symptoms.39


Affected tracts

Our systematic review confirmed that both descending corticomotor and ascending sensorimotor tracts are involved in the pathogenesis of CP and both are clinically significant (Fig. 2 outlines studies for and against involvement of particular tracts). It is not yet possible to say which is more significant; however, descending corticomotor tracts have been studied more comprehensively. Evidence for commissural and association fibre involvement is conflicting, most likely because of the small number of studies published so far that target these pathways. Further studies are warranted to elucidate fully how these networks are impacted upon in CP.

Figure 2.

 Evidence for and against involvement of ascending, descending, and other tracts in cerebral palsy.

The CST carries fibres from the motor cortex to the spinal cord, and was first suggested in 1962 to have involvement in spasticity.28 It is no surprise that with the advent of diffusion imaging the CST is the most frequently assessed tract in children with CP. Results repeatedly showed decreased fractional anisotropy and increased mean diffusivity within this white matter pathway. Such consistent findings suggest a decrease in the integrity of the CST compared with typically developing children. Further evidence of injury or perturbed early development of this pathway, namely reduced volume and decreased fibre count, has been demonstrated by dMRI and tractography studies.12,39,42,45,46,51 The prognostic use of this information is highlighted by the abnormal results preceding the clinical outcomes,39 and interestingly the compensatory changes after physical therapy and peripheral muscle botulinum toxin A administration.52 Other important descending corticomotor tracts are the corticobulbar tract, which also shows reduced fractional anisotropy and increased mean diffusivity, and regions within the PLIC which show reductions in fractional anisotropy, and normal or increased mean diffusivity.

The findings for the ascending sensorimotor pathways were more varied. The PTR was investigated in several studies as a sensory tract. The PTR connects the thalamus to the posterior parietal and occipital cortices. The posterior parietal cortex is involved with complex upper limb function and visuospatial performance,53 consistent with the finding that damage to the PTR correlates with GMFCS level and both motor and sensory function. The superior thalamic radiations (STR) connects the thalamus to the somatosensory cortex. This pathway was shown to have increased mean diffusivity as well as decreased fractional anisotropy only in severe cases. One recent study also showed statistically significant correlations between this pathway and upper limb function in spastic hemiplegia,12 which were not present in the CST. Combined, these findings highlight the importance of preservation of ascending sensorimotor networks in motor function, and provide new insights for the design of new neurorehabilitation therapies, which may enhance sensory pathways.

The evidence for involvement of the corpus callosum is conflicting, but suggests involvement with a similar pattern of reduced fractional anisotropy and increased mean diffusivity, correlated with clinical severity. The transcallosal motor fibres in particular have been shown to be involved in participants with spastic diplegia. Further research is warranted in this area.

Correlations with clinical measures

Over half the included studies assessed correlations between clinical measures and imaging parameters. The most frequently used assessment to compare with imaging indices was the GMFCS. Although this was shown to correlate with corticomotor, sensorimotor, and commissural tracts, interpretation of these results is difficult, as it is a classification system based on five categories, not a continuous scale. A structure–function correlation cannot be established using this measure, rather a comparison between groups. There were fewer studies that used clinical measurements with continuous variables, and among those that did, there was little consistency across studies. Use of standardized continuous clinical measures with well-documented reproducibility, such as the Melbourne Unimanual Upper Limb assessment of unimanual capacity,54 and Assisting Hand Assessment of bimanual coordination,55 would enhance interpretation of multiple studies and allow for meta-analyses in the future.

Differences between subgroups

Only two of the included studies have looked at differences between different subtypes of CP.48,49 Both studies have shown significant differences in both fractional anisotropy and mean diffusivity in specific tracts between different CP groups. Children with athetosis showed similar but more exaggerated changes than those with spasticity. Within the spasticity subtype of CP, children with quadriplegia showed changes in specific parts of the CST that were not evident in children with diplegia. These studies highlight the need to reduce heterogeneity of CP groups within dMRI studies in order to allow highlighting of specific changes and locations, which may correlate with the pathogenesis of each individual subtype. Few studies have assessed children with athetosis, as dystonia is infrequent, and further dMRI studies looking at correlations with clinical measures are likely to reveal further insight into the pathogenesis of this condition.

Study design

Study design of diffusion imaging studies in the research domain is extremely important to allow meaningful interpretation of results. Owing to the heterogeneity of acquisition parameters and analysis techniques, results are not directly comparable between studies. For a result to be meaningful, each study must compare ‘affected’ regions with some sort of control. Most easily interpreted is a comparison with typically developing controls; however, comparisons between ipsilesional and contralesional hemispheres in children with hemiplegia, or between different subtypes of CP, have also been used.

Limitations and methodology

The manual definition of ROIs for the analysis of diffusion anisotropy indices has limitations due to the high operator dependency. Only 35% of included studies using manual ROI placement included reliability analysis. Therefore, both the accuracy and repeatability of the remaining studies may be compromised. Additionally, the use of qualitative and semi-qualitative measurements for reporting used in some studies adds a second element of operator dependence. Ideally, quantitative metrics, such as streamline count, fractional anisotropy, and mean diffusivity should be reported, such that repeatability is maximized. Two included recent manuscripts have reported an automated approach to define ROIs.44,45 The major focus of these studies was to introduce an atlas-based framework for investigating brain pathology in CP. Such an approach has significant potential for more in-depth assessment of important structure–function relationships. Automated whole-brain analyses are essential to investigate neural networks in CP other than corticomotor pathways, such as those associated with executive function, which has been shown to be impaired in children with CP in recent neuropsychological studies.56–58 Neither automated parcellation nor manual ROI placement take individual variation in functional anatomy into account. To account for these, some recent studies (not involving CP) have seeded tractography studies using regions localized to each individual participant with functional MRI59 or transcranial magnetic stimulation.60

The conflicting findings in some corticomotor regions are likely to be a reflection of the heterogeneous natures of both the underlying pathology and the neuroimaging methodologies used. Ideally, studies with large numbers of participants with specific subtypes of CP could provide more convincing evidence for specific tract involvement. The acceptance of standardized acquisition imaging protocols for dMRI would also progressively build knowledge and further our understanding of CP pathology. Analysis methods such as those introduced by Faria et al.44,45 would help establish a more generalized framework for analysis of dMRI data.

A significant proportion of the dMRI studies so far have made use of anisotropy information gained from using the tensor model. A limitation of the tensor model is its inability to resolve crossing fibres in complex white matter populations,27 which has significant ramifications for the accuracy of tractography-derived diffusivity measures. This presents a significant challenge in dMRI studies as it has been estimated that crossing fibres are present in at least two-thirds of brain voxels with fractional anisotropy >0.2.61 Within these anatomical locations, the clinical interpretation of diffusivity measures is difficult, as voxels containing highly organized crossed white matter networks can exhibit paradoxical reduced anisotropy.27 Higher-order models of diffusion have been developed to overcome this limitation and will provide improved insight into white matter injury and neural reorganization in CP.

Although dMRI studies play a useful role in clinical management of CP,62 the full potential of this technology remains largely in the research domain. When sMRI is already indicated, adding dMRI sequences to the scan will add only a few minutes to the overall scan time. The major limitation is that for any quantitative analysis of the data, including tractography, which is mandatory for any connectivity analysis, offline analysis needs to be performed. As highlighted by the studies reviewed here, the preferred methodology for offline analysis is heterogeneous, and as yet, there are no standards by which results can be reliably interpreted.


Diffusion imaging studies in CP are providing new insight into the specific injury and reorganization of white matter motor pathways. Small sample sizes and heterogeneous imaging acquisition and analysis strategies have affected the clinical interpretation of findings. Given this constraint, there is corroborating evidence showing that decreased fractional anisotropy and increased mean diffusivity within descending corticomotor tracts, particularly the CST, are useful measures of white matter tract integrity, which correlate with measures of clinical severity of CP. There is also evidence to suggest that diffusion changes in ascending sensorimotor tracts, in particular the PTR, might provide novel information about corticomotor reorganization in CP. The link between these findings and motor function has been established, but less thoroughly investigated in larger clinical populations. Evidence for involvement of commissural and association fibres is limited and conflicting. There are no data on involvement of frontal, temporal, and occipital lobes. Although spasticity in varying motor distributions is well studied, children with athetosis are generally undersampled, and have been shown to have significantly different diffusion properties from those with spasticity in many brain regions. It may be of value for future studies to consider these types of CP separately.