Diffusion MRI of white matter alterations in chronic traumatic brain injury: a systematic review and meta-analysis☆ Authors: Yining Chen, Jessica Hoffmann, Warda T. Syeda, Remika Mito, Bradford A. Moffat Published: 2026 ABSTRACT BackgroundTraumatic brain injury (TBI) in the chronic stage (≥ 3 months post-injury) is associated with long-term symptoms and secondary conditions, which have been linked to persistent white matter alterations. Although these changes can be detected using diffusion magnetic resonance imaging (dMRI), it remains unclear whether the findings are consistently reported in the chronic phase of injury. This review aims to summarise and quantify differences in diffusion metrics between individuals with chronic non-sports/non-combat-related TBI and controls.MethodsThis review followed PRISMA guidelines. Database searches were conducted in MEDLINE, Embase, PsycINFO, and Scopus to identify studies examining dMRI metric changes in chronic TBI. Fractional anisotropy (FA) and mean diffusivity (MD) values were extracted to compute the standardised mean difference for individual fibre pathways. Meta-regression was performed to investigate the effect of clinical-demographic and imaging variables on FA effect sizes.ResultsTwenty-seven studies were included in the meta-analysis. Decreases in FA and increases in MD were found across multiple white matter pathways in chronic TBI, with more pronounced effects in moderate-to-severe TBI compared with mild injury. The overall largest and most consistent effects were observed in the corpus callosum. Changes in FA were influenced by age, sex, and magnetic field strength.ConclusionThis meta-analysis provides quantitative evidence of widespread white matter alterations in the chronic stage of TBI. dMRI measures (FA and MD) are sensitive biomarkers of white matter disruptions and exhibit changes associated with injury severity. Future research should carefully control for confounding factors to enhance data comparability across studies. Introduction Traumatic brain injury (TBI) is a brain injury caused by an external force, which can lead to temporary or permanent disruption of normal brain structure and function. It represents a major cause of morbidity and mortality across all age groups, with more than 60 million people sustaining a TBI worldwide each year (Dewan et al., 2018). The severity of TBI is classified as mild, moderate, and severe, based on a combination of clinical features and neuroimaging findings (Blennow et al., 2016). While some individuals experience full functional recovery within days or weeks of sustaining a TBI (i.e., the acute/subacute stage), others continue to experience ongoing symptoms for more than 3 months after the initial injury (i.e., the chronic phase) (Blennow et al., 2016; Haarbauer-Krupa et al., 2021; Nelson et al., 2023). These chronic sequelae encompass a wide range of neuropsychological impairments (Perry et al., 2016; Wilson et al., 2017), as well as an increased incidence of secondary conditions (Chen et al., 2011; Crane et al., 2016; Gardner et al., 2018). Even mild injuries, which account for most TBI events (80–90%), can result in prolonged functional limitations (McInnes et al., 2017; Nelson et al., 2019). Despite the impact of these persistent symptoms on patient wellbeing, the management of TBI is largely focused on the acute or subacute phases, possibly due to a lack of understanding of the pathophysiological mechanisms underlying these chronic consequences. However, the involvement of white matter alterations has been consistently reported over the past two decades, highlighting their potential role in mediating and predicting long-term functional outcomes of TBI (Bramlett and Dietrich, 2015; Jolly et al., 2021; Kinnunen et al., 2011; Smith et al., 2003). White matter injury in TBI typically occurs due to rapid acceleration or deceleration of the head at the moment of trauma, subjecting brain tissues to shearing and strain forces. This can directly cause deformation of white matter and lead to diffuse axonal injury (DAI), one of the most important pathological substrates of TBI (Smith et al., 2003). Midline white matter tracts are thought to be particularly vulnerable to injury induced by such mechanical forces, including the corpus callosum (CC), internal capsule, and those within the brainstem (Johnson et al., 2013). Moreover, axonal damage may persist or slowly progress over time, giving rise to a cascade of secondary injury processes that eventually result in axonal degeneration and tissue atrophy (Cole et al., 2018; Johnson et al., 2013; Meythaler et al., 2001). In individuals with TBI, the presence and extent of DAI have been considered strong predictors of neurodegeneration, brain volume loss, and poorer neurological and functional outcomes (Graham et al., 2020; van Eijck et al., 2018; Walker and Tesco, 2013). Owing to the microscopic and diffuse nature of traumatic axonal injury, its clinical diagnosis remains difficult, particularly in the absence of macrostructural changes that can be discerned by conventional neuroimaging tools. To overcome these limitations, diffusion magnetic resonance imaging (dMRI) has emerged as a promising technique that allows for non-invasive, in vivo visualisation of white matter changes (Hulkower et al., 2013). The contrast mechanism in dMRI arises from the random displacement of water molecules, which is restricted and hindered in biological tissues (e.g., brain) comprising multiple highly organised compartments (e.g., axonal fibres) (Le Bihan, 2003). As such, any alterations in the underlying white matter microstructure, whether due to physiological or injury-related pathological processes, will be reflected as net changes in quantitative measures derived from diffusion models. Diffusion tensor imaging (DTI) is one of the most widely implemented approaches that provides two key measures of diffusion properties: mean diffusivity (MD) and fractional anisotropy (FA) (Basser et al., 1994). MD, also known as the apparent diffusion coefficient (ADC), describes the overall magnitude of water displacement (Alexander et al., 2007). On the other hand, FA represents the degree of diffusion directional dependence, with its value ranging from 0 (perfect isotropic) to 1 (complete anisotropic) (Mori and Zhang, 2006). Within normal-appearing white matter, low MD and high FA values reflect limited overall water mobility and restricted diffusion along the organised and intact axonal bundles (Alexander et al., 2007). By contrast, in individuals with TBI, higher MD and lower FA are often reported, indicating disruptions in the underlying white matter architecture due to injury (Niogi and Mukherjee, 2010). Since the introduction of diffusion modelling, numerous studies have utilised DTI and quantitative diffusion measures to probe white matter alterations following TBI. However, the observed changes in FA and MD and their spatial patterns often appear heterogeneous across the literature (Hulkower et al., 2013; Lindsey et al., 2023), particularly in mild TBI. This could be due to relatively small sample sizes in individual studies, as well as variations in sample characteristics and methodological designs. A number of systematic reviews (Charney et al., 2026; Lindsey et al., 2023) and meta-analyses (Aoki and Inokuchi, 2016; Aoki et al., 2012; Wallace et al., 2018) have therefore been conducted to synthesise dMRI findings across different post-injury time points and subpopulations. While qualitative reviews from these studies have revealed FA and MD changes persisting in the long term following TBI, quantitative data syntheses focusing explicitly on the chronic phase of injury are not available. Given that primary and secondary pathophysiological processes of TBI may contribute to opposing changes in FA and MD (Kou et al., 2010; Shenton et al., 2012), it is necessary to examine diffusion metrics specifically within the chronic phase. Moreover, existing meta-analyses have mostly included either mixed or specific populations, such as sports- or combat-related injury cohorts, which may not be representative of the general population. To this end, a systematic review and meta-analysis of diffusion metric findings in individuals with non-sports- and non-combat-related chronic TBI is warranted to better identify convergence among existing evidence, provide clearer interpretation of the results, and address potential sources of heterogeneity. In this review, we aimed to synthesise quantitative differences in dMRI metrics, FA and MD, between individuals with chronic TBI and controls, and to estimate overall effect sizes for individual white matter tracts using a region-of-interest (ROI)-based approach. We then conducted meta-regression analyses to examine the relationships between estimated effect sizes and potential moderators, including clinical-demographic characteristics and MRI acquisition parameters, considering that these factors may impact the dMRI measurements. We hypothesised that FA and MD would be sensitive to chronic white matter changes, as indicated by lower FA and higher MD in individuals with chronic TBI relative to controls. Methods This study was conducted in accordance with the Preferred Reporting Items for Systematic reviews and Meta-Analyses (PRISMA) statement (Page et al., 2021a, Page et al., 2021b). A review protocol was registered with the International Prospective Register of Systematic Reviews (PROSPERO) under the registration number: CRD42024558142. Protocol amendment Two minor amendments were made to the original protocol. Firstly, the meta-analyses of additional diffusion metrics derived from other advanced models (e.g., diffusion kurtosis imaging (DKI) and neurite orientation dispersion and density imaging (NODDI)) were removed. During the initial screening process, only three DKI/NODDI studies met all inclusion criteria and reported the necessary diffusion measures for data synthesis. Given this limited number of studies, the current meta-analysis focused on DTI measures only. Secondly, also due to limited data, a quantitative synthesis of associations between clinical/neurocognitive outcomes and dMRI metrics was not undertaken; instead, a qualitative review was conducted. Search strategy Literature searches were performed using the following electronic databases: MEDLINE, Embase, PsycINFO, and Scopus. The main search terms were “diffusion magnetic resonance imaging” and “chronic traumatic brain injury”, along with abbreviations and relevant keywords. Search results were restricted to studies published in English language between 2004 and 2024. An example of the full search strategy is provided in the Supplementary Material. The initial search was conducted on 28th May 2024, and was repeated on 19th Aug 2025 to include more recently published records. Study selection and inclusion criteria Studies identified from database searches were screened in Covidence (Veritas Health Innovation, Melbourne, Australia). Duplicated records were removed either manually or automatically by Covidence before the formal screening process began. First, two reviewers (YC and JH) independently screened all records based on titles and abstracts to include potentially eligible studies. Next, the full texts of the studies were retrieved and independently screened by two groups of reviewers (group 1: YC; group 2: WTS, RM, and BAM) for final inclusion. Conflicts were resolved through discussion among the reviewers until consensus was reached on all decisions. Studies were included in the systematic review if they met all the following inclusion criteria: (1) full-length, peer-reviewed observational studies; (2) inclusion of adult participants (≥ 18 years) who sustained non-penetrating TBI and were in the chronic stage of injury (≥ 3 months post-injury); (3) inclusion of control participants without a history of TBI; (4) use of dMRI to study white matter changes; (5) studies that reported exact FA and/or MD values for both TBI and control groups based on ROI or whole-brain analysis. The exclusion criteria for the systematic review were as follows: (1) studies that were not full-length, observational, or peer-reviewed (e.g., reviews, case studies, and conference abstracts); (2) inclusion of children (< 18 years) or participants with blast- or sports-related injury; (3) studies without a control group; (4) studies that did not use dMRI; (5) studies that did not report FA and MD values. Samples included in the current systematic review and meta-analysis are representative of individuals with chronic TBI from the general population. We excluded studies involving sports-related injury, considering the risk of repeated exposure to concussion or sub-concussion within this population, and because recent reviews focusing specifically on this injury cohort have been published elsewhere (Hellewell et al., 2020; Koerte et al., 2023). Studies involving blast-related injury were also excluded due to potential confounding factors, such as the high prevalence of co-existing psychological conditions in military populations, which may reduce comparability with the general population (Chapman and Diaz-Arrastia, 2014). If study results were partially reported but all other inclusion criteria were met, the corresponding authors were contacted via email to obtain the missing data. One group of authors were contacted for missing data, but no additional information was obtained. In addition, for studies to be included in the meta-analysis, they must have provided complete statistical information required to calculate effect sizes (e.g., means and standard deviations or standard errors, or Cohen's d). FA and MD values were quality-checked prior to inclusion to ensure consistency of the reported outcomes. Data extraction Data from studies included in the meta-analysis were manually extracted by one reviewer (YC) and checked at random by three other reviewers (WTS, RM, and BAM). The information collected included clinical and demographic characteristics (sample size, age, sex, time since injury, single or multiple TBI, severity of injury, Glasgow Coma Scale (GCS) score, and cause of injury), MRI acquisition parameters and processing methods (scanner vendor, field strength, b-value, number of diffusion directions, repetition time (TR), echo time (TE), voxel size, processing software, analytic method, ROI definition, and dMRI metric), and outcome measures (means with standard deviations/standard errors or 95% confidence intervals of FA and/or MD). In cases where graphical displays were provided without numerical data, WebPlotDigitizer (version 5.2) (Rohatgi, 2024) was used to extract values from the graph. For studies including multiple follow-up time points, only data from the furthest time point after injury were extracted to best characterise the chronic effects of TBI. If FA and MD values were reported for the left and right hemispheres separately, the values were averaged to generate a single mean and standard deviation for the corresponding ROI, as in a previous meta-analysis (Wallace et al., 2018). Risk of bias assessment The included studies were assessed for risk of bias by one reviewer (YC) using the Risk Of Bias In Non-randomised Studies - of Exposures (ROBINS-E) tool (Higgins et al., 2024). The risk of bias for each study was addressed by seven domains, and an overall judgement was made based on the domain with the greatest risk of bias. Studies were rated as having low, moderate, high, or very high risk of bias, and those identified as high or very high risk were further excluded from the meta-analysis. Data synthesis All meta-analyses in the current review were performed with R (version 4.4.2), using the “meta” and “metafor” packages (Balduzzi et al., 2019; Viechtbauer, 2010). First, the standardised mean difference (SMD) Hedges' g (Hedges, 1981) was computed for FA and MD in individual white matter ROIs comparing individuals with chronic TBI and controls. Absolute Hedges' g values of 0.2, 0.5, and 0.8 were interpreted as small, moderate, and large effects, respectively. Analyses were carried out only for ROIs reported by at least two studies. To account for study heterogeneity, a random-effects model was applied when estimating overall effect sizes. We conducted separate meta-analyses for mild and moderate-to-severe TBI cohorts, considering that different injury severities may lead to distinct dMRI findings. Moderate and severe TBI were combined because only a limited number of studies reported them separately. The I2 statistics and 95% prediction intervals were calculated to quantify the inconsistency in effect sizes across studies. Publication bias was assessed by visual inspection of the funnel plot, in which asymmetry was indicative of potential publication bias. This visual inspection was further confirmed by Egger's test to estimate the degree and possibility of asymmetry in the funnel plot (Egger et al., 1997). Then, mixed-effects meta-regression analyses were conducted to evaluate the effects of age, sex (% male), post-injury interval, magnetic field strength, b-value, and the number of diffusion directions on FA for mild and moderate-to-severe TBI. Considering the relatively small number of studies included for most ROIs, we ran meta-regression analyses only for the genu, body, and splenium of the CC, which were the most frequently investigated regions. Finally, we performed sensitivity analyses to assess the robustness of our meta-analysis to specific methodological decisions. Given that chronic TBI may be defined using a time point later than 3 months post-injury, we conducted a sensitivity analysis restricted to studies in which participants were at least 6 months post-TBI. We also performed a sensitivity analysis including only studies with sample sizes greater than 10 for both the chronic TBI and control groups, considering the potential effects of very small studies on the overall effect estimates. Due to the limited number of studies reporting MD outcomes, meta-regression and sensitivity analyses were conducted only for FA. In addition, to improve the readability of the forest plots and allow visual comparisons across different fibre pathway groups, we classified white matter tracts into four major groups: association, commissural, projection, and brainstem tracts. For visualisation purposes, example fibre tracts were generated using TractSeg (Wasserthal et al., 2018) on a template constructed from healthy subject dMRI data from our prior work (Mito et al., 2024). Results A total of 7074 non-duplicate records were identified through two database searches and underwent screening, as shown in Fig. 1. Among these, 27 studies were finally included in the meta-analysis, comprising 682 individuals with non-sports- and non-combat-related chronic TBI and 744 control participants. Three studies reported on both mild and moderate-to-severe injury cohorts, thus yielding 14 records for mild TBI and 16 for moderate-to-severe TBI.Fig. 1PRISMA systematic review flow diagram depicting the identification, screening, and inclusion of studies. Adapted from Page et al. (2021a).Fig. 1 PRISMA systematic review flow diagram depicting the identification, screening, and inclusion of studies. Adapted from Page et al. (2021a). Study characteristics The clinical-demographic and dMRI acquisition details of studies included in the meta-analysis are summarised in Table 1, Table 2, respectively. The mean age of individuals with chronic TBI ranged from 22.0 to 53.2 years, and the proportion of females varied from 0 to 68.2% across studies. The average time since injury was 3 months to 12.8 years. Motor vehicle accidents were the most common cause of injury, while other causes included falls, assault, falling objects, and other. Most studies recruited individuals with a single TBI where specified.Table 1Clinical and demographic information of studies included in the meta-analysis.StudyGroupNo. of subjectsAge (y)Sex (M/F)Time since injury (y)Single vs. Multiple TBISeverity of injuryGCSCause of injuryMild TBIInglese et al., 2005TBI26n/sn/s5.7n/sMildn/sn/sHC293515/14Kraus et al., 2007*TBI2035.85 ± 9.398/127.71 ± 6.93n/sMildn/sMVA, bicycle, blunt head trauma, fallHC1832.83 ± 10.647/11Rutgers et al., 2008TBI12n/sn/s2.08n/sMildn/sn/sHC1037 ± 97/3Lo et al., 2009TBI1038.16 ± 8.615/57.38 ± 2.66n/sMildn/sMVA (80%), fall (20%)HC1044 ± 10.95/5Grossman et al., 2012TBI1540.8 ± 11.919/63.94 ± 2.46SingleMild14.87 ± 0.35MVA (40%), fall (20%), assault (27%), falling object (13%)HC1436.5 ± 12.39/5Kasahara et al., 2012*TBI1039.1 ± 11.417/35.6 ± 2.80n/sMild14.4 ± 0.70MVA (100%)HC1223 ± 4.6910/2Kim et al., 2015TBI2247.25 ± 10.917/150.996 ± 0.696SingleMild14.73 ± 0.62MVA (82%), bicycle (14%), falling object (4%)HC2145.05 ± 14.028/13Veeramuthu et al., 2015TBI30n/sn/s0.5 ± 0.01SingleMildn/sMVA (100%)HC1929.05 ± 5.8315/4Maruta et al., 2016TBI3234.5 ± 13.815/171.67 ± 1.09SingleMildn/sn/sHC12634.8 ± 13.515/17Jang and Lee, 2017TBI544.21 ± 5.263/22.54 ± 0.86SingleMild15 ± 0.00MVA (100%)HC645.22 ± 4.193/3Næss-Schmidt et al., 2017TBI2727.6 ± 6.411/160.27 ± 0.02n/sMildn/sMVA, fall, assault, otherHC2727.4 ± 6.211/16Chong et al., 2019TBI4937.7 ± 10.632/1710.0 ± 8.1n/sMildn/sn/sHC4138.1 ± 10.222/19Mohammadian et al., 2020TBI8547 ± 2059/260.69n/sMild14.68 ± 0.54MVA (32%), fall (51%), assault (9%), other (8%)OC3050 ± 2014/16Zane et al., 2021*TBI2031.05 ± 7.1220/04.10 ± 2.32n/sMildn/sMVA, fall, assault, otherHC2734.48 ± 8.8627/0Moderate to severe TBIKraus et al., 2007*TBI1734.88 ± 11.628/910.36 ± 7.95n/sModerate & severen/sMVA, bicycle, blunt head trauma, fallHC1832.83 ± 10.647/11Xu et al., 2007TBI926.44 ± 4.859/04.33 ± 1.41n/sSevere5.56 ± 2.50MVA (89%), fall (11%)HC112611/0Sidaros et al., 2008TBI2332.4 ± 12.917/61SingleSeveren/sMVA, fall, assaultHC1431.2 ± 8.19/5Kennedy et al., 2009TBI839.07 ± 12.376/27.0 ± 8.6n/sSeveren/sMVA (100%)HC840.11 ± 14.695/3Kumar et al., 2010TBI1635.25 ± 10.288/8∼2SingleModerate10.8MVA (69%), fall from height (31%)HC1737.35 ± 9.3410/7Kasahara et al., 2012*TBI538.2 ± 9.234/15.2 ± 3.56n/sSeveren/sMVA (100%)HC1223 ± 4.6910/2Shah et al., 2012TBI1421.98 ± 4.6212/20.54 ± 0.03SingleSevere5.29MVA, otherHC1522.20 ± 5.9313/2Kourtidou et al., 2013TBI1422.0 ± 4.612/20.55 ± 0.03SingleSevere5.3 ± 1.50n/sHC1320.9 ± 5.111/2Palacios et al., 2013TBI2627.40 ± 5.1516/104.20 ± 1.14SingleSevere5.19 ± 1.70MVA (100%)HC2228.57 ± 6.7412/10Owens et al., 2017TBI2039.05 ± 16.4515/50.87n/sModerate & Severen/sMVA (85%), fall (10%), equestrian accident (5%)HC2033.45 ± 11.7212/8McDonald et al., 2018TBI1745.64 ± 13.2114/312.82 ± 8.83n/sSeveren/sMVA (59%), fall (29%), assault (12%)HC1744.29 ± 14.6214/3Rigon et al., 2019TBI3153.19 ± 13.5822/98.95 ± 10.22MixedModerate & Severen/sMVA (34%), fall (54%), assault (6%), other (6%)HC2448.29 ± 20.148/16Wallace et al., 2020bTBI3142.5 ± 15.926/50.61 ± 0.13SingleModerate & Severe8.1 ± 3.6MVA (26%), pedestrian (6%), fall (36%), assault (13%), bicycle (19%)HC + OC10639.2 ± 16.759/47Grassi et al., 2021TBI2029.6 ± 6.817/3∼1SingleModerate & severeMVA (80%), assault (5%), running over (15%)HC20n/sn/sZane et al., 2021*TBI2434.33 ± 8.3424/04.10 ± 2.32n/sModerate & severen/sMVA, fall, assault, otherHC2734.48 ± 8.8627/0Raukola-Lindblom et al., 2023TBI4435 ± 9.414/307.2 ± 7.5n/sModerate & Severen/sn/sHC6736 ± 1030/37Note: Studies marked with an asterisk (*) included both mild and moderate-to-severe TBI cohorts. Values are presented as mean ± standard deviation, except for those in italics, which are reported as medians. y, years; M, male; F, female; m, months; GCS, Glasgow Coma Scale; TBI, traumatic brain injury; HC, healthy control; n/s, not specified; MVA, motor vehicle accident; OC, orthopaedic control.Table 2Technical information of studies included in the meta-analysis.StudyScanner vendorField strength (T)b-value (s/mm2)Diffusion directionsTR (msec)TE (msec)Voxel size (mm3)Processing softwareAnalytic methodROI definitiondMRI metricMild TBIInglese et al., 2005Siemens Vision1.51000650001001.9 × 1.9 × 5.0In-house IDL softwareHistogram + ROIManually drawnFA & MDKraus et al., 2007*General Electric Signa VHi3.0750275200811.67 × 1.67 × 5.0DTI StudioROIManually drawnFARutgers et al., 2008Siemens Sonata1.510002557001101.875 × 1.875 × 4.0DPTools + MedINRIATractography + ROIManually drawnFA & ADCLo et al., 2009General Electric Signa Excite1.51000258700892.03 × 2.03 × 5.0FSLROIManually drawnFA & ADCGrossman et al., 2012Siemens Magnetom Trio3.010003021001151.7 × 1.7 × 3.0SPM + In-house MATLAB scriptsROIManually drawnFA & MDKasahara et al., 2012*Siemens Symphony1.51000605400991.875 × 1.875 × 3.3SPM8ROIAtlas-basedFAKim et al., 2015Philips Gyroscan Intera1.510003210,726762.5 × 2.5 × 2.5FSLTractography + ROIManually drawnFA & MDVeeramuthu et al., 2015General Electric Signa HDx3.07003213,00081.21.875 × 1.875 × 3.0FSLTBSS + ROIAtlas-basedFA & MDMaruta et al., 2016General Electric Signa Excite3.0100055n/sn/s0.8 × 0.8 × 1.8FSLTBSS + ROIAtlas-basedFAJang and Lee, 2017Philips Gyroscan Intera1.51000n/s10,726762.5 × 2.5 × 2.5DTI task cardTractography + ROIManually drawnFANæss-Schmidt et al., 2017Siemens Magnetom Skyra3.010003211,400792.3 × 2.3 × 2.3FSL + brainSuiteROIAutomated segmentationFA & MDChong et al., 2019Siemens Magnetom Skyra3.01000305100731.7 × 1.7 × 4.0FreeSurfer TRACULAROITractography-basedMDMohammadian et al., 2020Siemens Magnetom Verio3.010006411,7001062.0 × 2.0 × 2.0ExploreDTITBSS + ROIAtlas-basedFA & MDZane et al., 2021Philips Achieva3.0100015n/sn/s0.875 × 0.875 × 2.0FSLTBSS + ROIAtlas-basedFAModerate to severe TBIKraus et al., 2007*See aboveXu et al., 2007Philips Intera3.080032n/s501.80 × 1.80 × 1.72FSL + DtiStudioVoxel-based + ROI + TractographyManually drawnFA & ADCSidaros et al., 2008Siemens Magnetom Vision1.57406n/s601.80 × 1.80 × 5.0SPM2ROIManually drawnFA & MDKennedy et al., 2009Siemens Trio3.01000n/s8300862.0 × 2.0 × 2.0FSLROISemi-automated segmentationFA & MDKumar et al., 2010General Electric1.510001080001001.88 × 1.88 × 3.0JAVA-based programROIManually drawnFA & MDKasahara et al., 2012*See aboveShah et al., 2012Philips Intera3.0100030631851n/sPhilips toolROI + TractographyManually drawnFA & ADCKourtidou et al., 2013Philips Intera3.0100030631851n/sPhilips PRIDEROI + TractographyManually drawnFA & ADCPalacios et al., 2013Siemens Magnetom TrioTim3.01000309300942.0 × 2.0 × 2.0FSLWhole-brain + TBSSAutomated segmentationFAOwens et al., 2017Siemens Magnetom Skyra3.020006410,9001012.0 × 2.0 × 2.0FSLTractography + ROITractography-basedFA & MDMcDonald et al., 2018Philips Achieva3.01000618400682.5 × 2.5 × 2.5FSLROIManually drawnFA & MDRigon et al., 2019General Electric MR750W3.010003015,33974.52.0 × 2.0 × 2.0FSLTBSS + ROIAtlas-basedFA & MDWallace et al., 2020bSiemens Tim Trio3.030006494001162.5 × 2.5 × 2.5FSLROIAtlas-basedFA & MDGrassi et al., 2021Philips Intera Achieva3.01000328500612.0 × 2.0 × 2.0FSL + ExploreDTIROI + TractographyTractography-basedFA & MDZane et al., 2021*See aboveRaukola-Lindblom et al., 2023Philips Intera3.0800323347.683.81.75 × 1.75 × 2.0n/sROI + TractographyTractography-basedFANote: Studies marked with an asterisk (*) included both mild and moderate-to-severe TBI cohorts. T, tesla; TR, repetition time; TE, echo time; ROI, region of interest; dMRI, diffusion MRI; FA, fractional anisotropy; MD, mean diffusivity; ADC, apparent diffusion coefficient; TBSS, tract-based spatial statistics; n/s, not specified. Table 1Clinical and demographic information of studies included in the meta-analysis.StudyGroupNo. of subjectsAge (y)Sex (M/F)Time since injury (y)Single vs. Multiple TBISeverity of injuryGCSCause of injuryMild TBIInglese et al., 2005TBI26n/sn/s5.7n/sMildn/sn/sHC293515/14Kraus et al., 2007*TBI2035.85 ± 9.398/127.71 ± 6.93n/sMildn/sMVA, bicycle, blunt head trauma, fallHC1832.83 ± 10.647/11Rutgers et al., 2008TBI12n/sn/s2.08n/sMildn/sn/sHC1037 ± 97/3Lo et al., 2009TBI1038.16 ± 8.615/57.38 ± 2.66n/sMildn/sMVA (80%), fall (20%)HC1044 ± 10.95/5Grossman et al., 2012TBI1540.8 ± 11.919/63.94 ± 2.46SingleMild14.87 ± 0.35MVA (40%), fall (20%), assault (27%), falling object (13%)HC1436.5 ± 12.39/5Kasahara et al., 2012*TBI1039.1 ± 11.417/35.6 ± 2.80n/sMild14.4 ± 0.70MVA (100%)HC1223 ± 4.6910/2Kim et al., 2015TBI2247.25 ± 10.917/150.996 ± 0.696SingleMild14.73 ± 0.62MVA (82%), bicycle (14%), falling object (4%)HC2145.05 ± 14.028/13Veeramuthu et al., 2015TBI30n/sn/s0.5 ± 0.01SingleMildn/sMVA (100%)HC1929.05 ± 5.8315/4Maruta et al., 2016TBI3234.5 ± 13.815/171.67 ± 1.09SingleMildn/sn/sHC12634.8 ± 13.515/17Jang and Lee, 2017TBI544.21 ± 5.263/22.54 ± 0.86SingleMild15 ± 0.00MVA (100%)HC645.22 ± 4.193/3Næss-Schmidt et al., 2017TBI2727.6 ± 6.411/160.27 ± 0.02n/sMildn/sMVA, fall, assault, otherHC2727.4 ± 6.211/16Chong et al., 2019TBI4937.7 ± 10.632/1710.0 ± 8.1n/sMildn/sn/sHC4138.1 ± 10.222/19Mohammadian et al., 2020TBI8547 ± 2059/260.69n/sMild14.68 ± 0.54MVA (32%), fall (51%), assault (9%), other (8%)OC3050 ± 2014/16Zane et al., 2021*TBI2031.05 ± 7.1220/04.10 ± 2.32n/sMildn/sMVA, fall, assault, otherHC2734.48 ± 8.8627/0Moderate to severe TBIKraus et al., 2007*TBI1734.88 ± 11.628/910.36 ± 7.95n/sModerate & severen/sMVA, bicycle, blunt head trauma, fallHC1832.83 ± 10.647/11Xu et al., 2007TBI926.44 ± 4.859/04.33 ± 1.41n/sSevere5.56 ± 2.50MVA (89%), fall (11%)HC112611/0Sidaros et al., 2008TBI2332.4 ± 12.917/61SingleSeveren/sMVA, fall, assaultHC1431.2 ± 8.19/5Kennedy et al., 2009TBI839.07 ± 12.376/27.0 ± 8.6n/sSeveren/sMVA (100%)HC840.11 ± 14.695/3Kumar et al., 2010TBI1635.25 ± 10.288/8∼2SingleModerate10.8MVA (69%), fall from height (31%)HC1737.35 ± 9.3410/7Kasahara et al., 2012*TBI538.2 ± 9.234/15.2 ± 3.56n/sSeveren/sMVA (100%)HC1223 ± 4.6910/2Shah et al., 2012TBI1421.98 ± 4.6212/20.54 ± 0.03SingleSevere5.29MVA, otherHC1522.20 ± 5.9313/2Kourtidou et al., 2013TBI1422.0 ± 4.612/20.55 ± 0.03SingleSevere5.3 ± 1.50n/sHC1320.9 ± 5.111/2Palacios et al., 2013TBI2627.40 ± 5.1516/104.20 ± 1.14SingleSevere5.19 ± 1.70MVA (100%)HC2228.57 ± 6.7412/10Owens et al., 2017TBI2039.05 ± 16.4515/50.87n/sModerate & Severen/sMVA (85%), fall (10%), equestrian accident (5%)HC2033.45 ± 11.7212/8McDonald et al., 2018TBI1745.64 ± 13.2114/312.82 ± 8.83n/sSeveren/sMVA (59%), fall (29%), assault (12%)HC1744.29 ± 14.6214/3Rigon et al., 2019TBI3153.19 ± 13.5822/98.95 ± 10.22MixedModerate & Severen/sMVA (34%), fall (54%), assault (6%), other (6%)HC2448.29 ± 20.148/16Wallace et al., 2020bTBI3142.5 ± 15.926/50.61 ± 0.13SingleModerate & Severe8.1 ± 3.6MVA (26%), pedestrian (6%), fall (36%), assault (13%), bicycle (19%)HC + OC10639.2 ± 16.759/47Grassi et al., 2021TBI2029.6 ± 6.817/3∼1SingleModerate & severeMVA (80%), assault (5%), running over (15%)HC20n/sn/sZane et al., 2021*TBI2434.33 ± 8.3424/04.10 ± 2.32n/sModerate & severen/sMVA, fall, assault, otherHC2734.48 ± 8.8627/0Raukola-Lindblom et al., 2023TBI4435 ± 9.414/307.2 ± 7.5n/sModerate & Severen/sn/sHC6736 ± 1030/37Note: Studies marked with an asterisk (*) included both mild and moderate-to-severe TBI cohorts. Values are presented as mean ± standard deviation, except for those in italics, which are reported as medians. y, years; M, male; F, female; m, months; GCS, Glasgow Coma Scale; TBI, traumatic brain injury; HC, healthy control; n/s, not specified; MVA, motor vehicle accident; OC, orthopaedic control. Clinical and demographic information of studies included in the meta-analysis. Note: Studies marked with an asterisk (*) included both mild and moderate-to-severe TBI cohorts. Values are presented as mean ± standard deviation, except for those in italics, which are reported as medians. y, years; M, male; F, female; m, months; GCS, Glasgow Coma Scale; TBI, traumatic brain injury; HC, healthy control; n/s, not specified; MVA, motor vehicle accident; OC, orthopaedic control. Table 2Technical information of studies included in the meta-analysis.StudyScanner vendorField strength (T)b-value (s/mm2)Diffusion directionsTR (msec)TE (msec)Voxel size (mm3)Processing softwareAnalytic methodROI definitiondMRI metricMild TBIInglese et al., 2005Siemens Vision1.51000650001001.9 × 1.9 × 5.0In-house IDL softwareHistogram + ROIManually drawnFA & MDKraus et al., 2007*General Electric Signa VHi3.0750275200811.67 × 1.67 × 5.0DTI StudioROIManually drawnFARutgers et al., 2008Siemens Sonata1.510002557001101.875 × 1.875 × 4.0DPTools + MedINRIATractography + ROIManually drawnFA & ADCLo et al., 2009General Electric Signa Excite1.51000258700892.03 × 2.03 × 5.0FSLROIManually drawnFA & ADCGrossman et al., 2012Siemens Magnetom Trio3.010003021001151.7 × 1.7 × 3.0SPM + In-house MATLAB scriptsROIManually drawnFA & MDKasahara et al., 2012*Siemens Symphony1.51000605400991.875 × 1.875 × 3.3SPM8ROIAtlas-basedFAKim et al., 2015Philips Gyroscan Intera1.510003210,726762.5 × 2.5 × 2.5FSLTractography + ROIManually drawnFA & MDVeeramuthu et al., 2015General Electric Signa HDx3.07003213,00081.21.875 × 1.875 × 3.0FSLTBSS + ROIAtlas-basedFA & MDMaruta et al., 2016General Electric Signa Excite3.0100055n/sn/s0.8 × 0.8 × 1.8FSLTBSS + ROIAtlas-basedFAJang and Lee, 2017Philips Gyroscan Intera1.51000n/s10,726762.5 × 2.5 × 2.5DTI task cardTractography + ROIManually drawnFANæss-Schmidt et al., 2017Siemens Magnetom Skyra3.010003211,400792.3 × 2.3 × 2.3FSL + brainSuiteROIAutomated segmentationFA & MDChong et al., 2019Siemens Magnetom Skyra3.01000305100731.7 × 1.7 × 4.0FreeSurfer TRACULAROITractography-basedMDMohammadian et al., 2020Siemens Magnetom Verio3.010006411,7001062.0 × 2.0 × 2.0ExploreDTITBSS + ROIAtlas-basedFA & MDZane et al., 2021Philips Achieva3.0100015n/sn/s0.875 × 0.875 × 2.0FSLTBSS + ROIAtlas-basedFAModerate to severe TBIKraus et al., 2007*See aboveXu et al., 2007Philips Intera3.080032n/s501.80 × 1.80 × 1.72FSL + DtiStudioVoxel-based + ROI + TractographyManually drawnFA & ADCSidaros et al., 2008Siemens Magnetom Vision1.57406n/s601.80 × 1.80 × 5.0SPM2ROIManually drawnFA & MDKennedy et al., 2009Siemens Trio3.01000n/s8300862.0 × 2.0 × 2.0FSLROISemi-automated segmentationFA & MDKumar et al., 2010General Electric1.510001080001001.88 × 1.88 × 3.0JAVA-based programROIManually drawnFA & MDKasahara et al., 2012*See aboveShah et al., 2012Philips Intera3.0100030631851n/sPhilips toolROI + TractographyManually drawnFA & ADCKourtidou et al., 2013Philips Intera3.0100030631851n/sPhilips PRIDEROI + TractographyManually drawnFA & ADCPalacios et al., 2013Siemens Magnetom TrioTim3.01000309300942.0 × 2.0 × 2.0FSLWhole-brain + TBSSAutomated segmentationFAOwens et al., 2017Siemens Magnetom Skyra3.020006410,9001012.0 × 2.0 × 2.0FSLTractography + ROITractography-basedFA & MDMcDonald et al., 2018Philips Achieva3.01000618400682.5 × 2.5 × 2.5FSLROIManually drawnFA & MDRigon et al., 2019General Electric MR750W3.010003015,33974.52.0 × 2.0 × 2.0FSLTBSS + ROIAtlas-basedFA & MDWallace et al., 2020bSiemens Tim Trio3.030006494001162.5 × 2.5 × 2.5FSLROIAtlas-basedFA & MDGrassi et al., 2021Philips Intera Achieva3.01000328500612.0 × 2.0 × 2.0FSL + ExploreDTIROI + TractographyTractography-basedFA & MDZane et al., 2021*See aboveRaukola-Lindblom et al., 2023Philips Intera3.0800323347.683.81.75 × 1.75 × 2.0n/sROI + TractographyTractography-basedFANote: Studies marked with an asterisk (*) included both mild and moderate-to-severe TBI cohorts. T, tesla; TR, repetition time; TE, echo time; ROI, region of interest; dMRI, diffusion MRI; FA, fractional anisotropy; MD, mean diffusivity; ADC, apparent diffusion coefficient; TBSS, tract-based spatial statistics; n/s, not specified. Technical information of studies included in the meta-analysis. Note: Studies marked with an asterisk (*) included both mild and moderate-to-severe TBI cohorts. T, tesla; TR, repetition time; TE, echo time; ROI, region of interest; dMRI, diffusion MRI; FA, fractional anisotropy; MD, mean diffusivity; ADC, apparent diffusion coefficient; TBSS, tract-based spatial statistics; n/s, not specified. The imaging protocols used for dMRI acquisition varied considerably, such as the echo time (range = 50–116 msec), repetition time (range = 2100–15,330 msec), in-plane resolution (range = 0.8 × 0.8 to 2.5 × 2.5 mm2), slice thickness (range = 1.72–5.0 mm), b-value (range = 700–3000 s/mm2), and the number of diffusion-encoding directions (range = 6–64). A variety of software packages were used to process and analyse the diffusion data, among which the FMRIB's Software Library (FSL) (Jenkinson et al., 2012; Smith et al., 2004) was the most frequently used. Moreover, the definition of white matter ROIs was inconsistent across studies: ROIs were either manually drawn (48%), based on the ICBM-DTI-81 atlas (26%) (Mori et al., 2008), defined via tractography (15%), or automatically segmented (11%). Systematic review of dMRI findings in chronic TBI An overview of the classification of fibre pathways reported by at least two studies is illustrated in Fig. 2. For studies examining FA changes in chronic mild TBI, six identified significantly lower FA in various white matter ROIs compared to controls (Grossman et al., 2012; Inglese et al., 2005; Kraus et al., 2007; Lo et al., 2009; Mohammadian et al., 2020; Veeramuthu et al., 2015), with FA reductions in the superior longitudinal fasciculus, genu of the CC, and internal capsule observed in two studies each. By contrast, six studies did not identify any significant group differences in FA across the ROIs examined (Jang and Lee, 2017; Kasahara et al., 2012; Maruta et al., 2016; Næss-Schmidt et al., 2017; Rutgers et al., 2008; Zane et al., 2021). One study further divided individuals with TBI into two groups based on whether they experienced neuropathic pain, and significantly lower FA was observed only in the pain group compared with controls (Kim et al., 2015). Notably, one study reported significantly higher FA values in the internal capsule of individuals with chronic TBI (Lo et al., 2009).Fig. 2Classification of key white matter fibre pathways. Tracts are categorised into association fibres, commissural fibres, projection fibres, and tracts in the brainstem. For fibre tracts that are represented bilaterally, only the left side is shown in the figure. Of note, tracts are shown here for illustrative purposes and do not necessarily correspond spatially to the ROIs included in the reported studies. White matter fibre tracts were defined variably across studies included in the systematic review and meta-analysis.Fig. 2 Classification of key white matter fibre pathways. Tracts are categorised into association fibres, commissural fibres, projection fibres, and tracts in the brainstem. For fibre tracts that are represented bilaterally, only the left side is shown in the figure. Of note, tracts are shown here for illustrative purposes and do not necessarily correspond spatially to the ROIs included in the reported studies. White matter fibre tracts were defined variably across studies included in the systematic review and meta-analysis. Among the studies evaluating group differences in MD between individuals with mild TBI and controls, six reported significantly higher MD in the TBI group (Chong et al., 2019; Grossman et al., 2012; Inglese et al., 2005; Lo et al., 2009; Mohammadian et al., 2020; Veeramuthu et al., 2015), two found no significant group differences (Næss-Schmidt et al., 2017; Rutgers et al., 2008), and one observed significantly higher MD in patients with post-traumatic neuropathic pain but not in the non-pain TBI group (Kim et al., 2015). Significantly higher MD values were reported in more than one study in the internal capsule, cingulum, and splenium of the CC. Sixteen studies reported FA findings in chronic moderate-to-severe TBI, and all identified significantly lower FA in the TBI group across multiple white matter ROIs (Grassi et al., 2021; Kasahara et al., 2012; Kennedy et al., 2009; Kourtidou et al., 2013; Kraus et al., 2007; Kumar et al., 2010; McDonald et al., 2018; Owens et al., 2017; Palacios et al., 2013; Raukola-Lindblom et al., 2023; Rigon et al., 2019; Shah et al., 2012; Sidaros et al., 2008; Wallace et al., 2020b; Xu et al., 2007; Zane et al., 2021). Lower FA was most frequently observed in the splenium of the CC, followed by the genu and body of the CC, and the superior longitudinal fasciculus. In moderate-to-severe TBI studies, the selection of ROIs was often based on specific a priori hypotheses, thus resulting in some regions with lower FA being reported by a single study only, such as fibre pathways through the ventral striatum (Shah et al., 2012), supero-lateral branch of medial forebrain bundle (Owens et al., 2017), and frontal and temporal connections of the lower parietal cortex (Raukola-Lindblom et al., 2023). In addition, three studies also included a mild TBI cohort for comparison across injury severities (Kasahara et al., 2012; Kraus et al., 2007; Zane et al., 2021). These studies showed that significantly lower FA was found in more ROIs in moderate-to-severe TBI, whereas FA values in individuals with mild TBI more closely resembled that of controls. In moderate-to-severe TBI studies, eleven reported group differences in MD between individuals with TBI and controls, all of which identified significantly higher MD in various white matter ROIs (Grassi et al., 2021; Kennedy et al., 2009; Kourtidou et al., 2013; Kumar et al., 2010; McDonald et al., 2018; Owens et al., 2017; Rigon et al., 2019; Shah et al., 2012; Sidaros et al., 2008; Wallace et al., 2020b; Xu et al., 2007). Again, higher MD values were most commonly found in the splenium and body of the CC. Relationship between diffusion metrics and clinical findings Several studies investigated associations between diffusion metrics in specific ROIs and measures of neurocognitive performance across different domains (e.g., attention, memory, executive functioning, processing speed, language). In mild TBI, three studies reported the extent to which FA and/or MD were associated with some aspects of neurocognitive performance. One study found positive correlations with FA (i.e., higher FA was associated with improved neurocognitive functioning) (Kraus et al., 2007), one study showed both positive and negative correlations with FA but no association with MD (Veeramuthu et al., 2015), and one study reported no correlations with either metric (Grossman et al., 2012). In moderate-to-severe TBI, nine out of 16 studies examined the relationship between diffusion metrics and neurocognitive functioning. One study reported positive correlations with FA (Zane et al., 2021), while one study noted negative correlations with FA (Raukola-Lindblom et al., 2023); two studies showed positive correlations with FA while no association with MD (Grassi et al., 2021; McDonald et al., 2018), two studies revealed positive correlations with FA and negative correlations with MD (i.e., lower MD was associated with better neurocognitive performance) (Kourtidou et al., 2013; Rigon et al., 2019), and three studies found no correlations with either FA or MD (Kumar et al., 2010; Owens et al., 2017; Wallace et al., 2020b). Similarly, two studies explored associations between diffusion metrics in selected ROIs and other clinical findings. One study found no significant correlation between FA and post-concussion symptoms in mild TBI (Næss-Schmidt et al., 2017), whereas another study reported positive correlations between FA and Glasgow Outcome Scale scores (i.e., higher FA was associated with improved functional outcomes) (Sidaros et al., 2008). Meta-analyses of FA and MD in chronic TBI The full results of the meta-analyses for all white matter ROIs are provided in the Supplementary Material (Tables S2-S5). In individuals with mild TBI, meta-analyses revealed small-to-moderate FA reductions in 87% of the 23 white matter ROIs examined (Fig. 3a), although significant group differences were only found in the uncinate fasciculus (Hedges' g = −0.33, 95% CI = −0.64 to −0.01, p = 0.040, I2 = 15.9%) and centrum semiovale (Hedges' g = −0.50, 95% CI = −0.94 to −0.06, p = 0.024, I2 = 0%). Eight ROIs had estimable effect sizes for MD, all of which showed higher MD in individuals with mild TBI compared with controls (Fig. 4a). Statistically significant MD increases were observed in the splenium of the CC (Hedges' g = 0.64, 95% CI = 0.27–1.02, p = 0.001, I2 = 34.6%), posterior limb of the internal capsule (Hedges' g = 0.56, 95% CI = 0.18–0.94, p = 0.004, I2 = 39.4%), and middle cerebellar peduncle (Hedges' g = 0.66, 95% CI = 0.31–1.00, p < 0.001, I2 = 0%).Fig. 3Summary forest plots of Hedges' g effect sizes for FA in (a) mild and (b) moderate-to-severe TBI. Negative effect sizes (g < 0) indicate lower FA values in individuals with chronic TBI relative to controls. CI, confidence interval; NStudy, number of studies included; NTBI/Control, number of individuals with TBI/controls.Fig. 3Fig. 4Summary forest plots of Hedges' g effect sizes for MD in (a) mild and (b) moderate-to-severe TBI. Positive effect sizes (g > 0) indicate higher MD values in individuals with chronic TBI relative to controls. CI, confidence interval. NStudy, number of studies included; NTBI/Control, number of individuals with TBI/controls.Fig. 4 Summary forest plots of Hedges' g effect sizes for FA in (a) mild and (b) moderate-to-severe TBI. Negative effect sizes (g < 0) indicate lower FA values in individuals with chronic TBI relative to controls. CI, confidence interval; NStudy, number of studies included; NTBI/Control, number of individuals with TBI/controls. Summary forest plots of Hedges' g effect sizes for MD in (a) mild and (b) moderate-to-severe TBI. Positive effect sizes (g > 0) indicate higher MD values in individuals with chronic TBI relative to controls. CI, confidence interval. NStudy, number of studies included; NTBI/Control, number of individuals with TBI/controls. In moderate-to-severe TBI, meta-analyses demonstrated overall reductions in FA (Fig. 3b) and increases in MD (Fig. 4b), with all white matter ROIs showing moderate-to-large effect sizes. Significant group differences for both FA and MD were observed in the superior longitudinal fasciculus (FA: Hedges' g = −0.68, 95% CI = −1.04 to −0.32, p < 0.001, I2 = 60.9%; MD: Hedges' g = 0.95, 95% CI = 0.60–1.30, p < 0.001, I2 = 0%), as well as in the body (FA: Hedges' g = −1.66, 95% CI = −2.18 to −1.14, p < 0.001, I2 = 70.3%; MD: Hedges' g = 1.29, 95% CI = 1.00–1.57, p < 0.001, I2 = 0%) and splenium (FA: Hedges' g = −1.47, 95% CI = −1.90 to −1.05, p < 0.001, I2 = 62.9%; MD: Hedges' g = 1.77, 95% CI = 0.77–2.78, p < 0.001, I2 = 81.8%) of the CC. Assessment of publication bias Given that funnel plots can be unreliable when fewer than 10 studies are included in a meta-analysis, as was the case for individual white matter ROIs in our study, publication bias was assessed by calculating pooled effect sizes for FA across all ROIs within each study. In mild TBI, the funnel plot appeared symmetrical upon visual inspection. Egger's test further confirmed no evidence of plot asymmetry (intercept = 0.08, 95% CI = −0.44 to 0.60, p = 0.258) (Fig. 5a). For moderate-to-severe TBI studies, the funnel plot appeared to be slightly skewed to the left, with smaller studies reporting larger negative effect sizes. Potential funnel plot asymmetry was indicated by Egger's test (intercept = 0.51, 95% CI = −0.32 to 1.33, p = 0.006) (Fig. 5b).Fig. 5Funnel plots for (a) mild and (b) moderate-to-severe TBI studies. Each point represents the FA effect size for an individual study plotted against its standard error. The dashed lines indicate 95% confidence intervals. Of note, only studies reporting ROIs included in the forest plots are shown; therefore, the number of study points is smaller than the total number of studies included in the meta-analysis.Fig. 5 Funnel plots for (a) mild and (b) moderate-to-severe TBI studies. Each point represents the FA effect size for an individual study plotted against its standard error. The dashed lines indicate 95% confidence intervals. Of note, only studies reporting ROIs included in the forest plots are shown; therefore, the number of study points is smaller than the total number of studies included in the meta-analysis. Meta-regression analyses In mild TBI, meta-regression analyses revealed that a higher proportion of male participants was significantly associated with greater FA reductions in the genu (p = 0.040), body (p = 0.009), and splenium (p = 0.033) of the CC. The effect estimate for FA in mild TBI was not influenced by other clinical-demographic or imaging variables (Supplementary Table S6). In moderate-to-severe TBI, reductions in FA diminished with advancing age in the body of the CC (p = 0.023), and higher field strength was associated with smaller FA reductions in the genu (p = 0.001) and splenium (p = 0.038) of the CC. No significant associations were found between FA effect sizes and sex, post-injury time, b-value, or the number of diffusion directions (Supplementary Table S7). Sensitivity analyses Studies included in the sensitivity analyses are listed in Supplementary Table S8. The sensitivity analysis considering only studies with participants who were at least 6 months post-TBI yielded findings consistent with our primary analysis (Supplementary Table S9 & Fig. S2). In the sensitivity analysis restricted to studies with a sample size >10 per group, the results were also highly consistent with the primary analysis, with the only exception being the internal capsule in mild TBI (Supplementary Table S10 & Fig. S4). Significantly lower FA was identified in the internal capsule in individuals with chronic TBI compared to controls (Hedges' g = −0.57, 95% CI = −1.01 to −0.14, p = 0.01, I2 = 0%), although this FA reduction was not significant in the primary analysis. This change was driven by the exclusion of a single small study (Lo et al., 2009), which reported significantly higher FA in the chronic TBI group relative to controls. Discussion In the present study, we summarised and synthesised dMRI findings from 27 studies examining regional FA and MD differences between individuals with chronic TBI from the general population and controls. Our meta-analyses identified widespread reductions in FA and increases in MD across multiple ROIs, which are indicative of diffuse white matter alterations persisting months to years after the initial trauma. Even mild injuries can lead to detectable changes in diffusion metrics, highlighting the sensitivity of DTI measures for capturing microstructural white matter abnormalities. The most frequently investigated ROI across injury severity was the splenium of the CC, followed by the genu, and body of the CC. In mild TBI, most white matter tracts exhibited small FA effect sizes, whereas in moderate-to-severe TBI, the effect sizes became markedly larger, reflecting a greater extent of white matter alterations. This finding also suggests a relationship between TBI injury severity and the degree of chronic white matter changes detected and quantified by dMRI metrics. Subsequent meta-regression analyses showed that FA effect sizes in the CC were associated with age, sex, and magnetic field strength. In contrast, the effects of post-injury interval, b-value, and the number of diffusion directions on FA changes were not significant. In our study inclusion criteria, chronic TBI was defined as at least 3 months post-injury, in line with some previous TBI reviews (Haarbauer-Krupa et al., 2021; Lindsey et al., 2023; McInnes et al., 2017). However, the definition of “chronic” is indeed variable across the literature, and longer post-injury intervals may reflect a more stable recovery stage. Sensitivity analyses were therefore performed to assess the robustness of our results, which were only possible for FA due to the limited number of studies reporting MD outcomes. In the sensitivity analysis restricted to studies with participants assessed at ≥ 6 months post-injury, the results were highly consistent with the primary meta-analysis. This suggests that our findings are robust and not strongly driven by studies including individuals in the early chronic phase of TBI. In the sensitivity analysis restricted to studies with sample sizes > 10, the FA effect size in the internal capsule became statistically significant, in contrast to the non-significant finding in the primary meta-analysis. This change was attributable to the exclusion of a single small study, indicating the potential influence of small-sample studies on pooled effect estimates. The impact of excluding small studies on the assessment of publication bias was further discussed in the Limitations section. We found some evidence that diffusion metrics in selected ROIs may be associated with clinical and neurocognitive outcomes in the chronic phase of TBI, although the findings are highly inconsistent across the studies considered in the review. For instance, both higher FA and lower MD were associated with improved neurocognitive performance in some studies (Kourtidou et al., 2013; Rigon et al., 2019), whereas other studies reported no significant associations (Grossman et al., 2012; Kumar et al., 2010; Owens et al., 2017; Wallace et al., 2020b), or negative associations between FA and neurocognitive outcomes (Raukola-Lindblom et al., 2023). This inconsistency may be due to variations not only in the definition of white matter ROIs, but also in the neurocognitive tests used and their relative sensitivity to specific domains of neurocognitive impairment. A quantitative synthesis with more available data is needed in future studies to better resolve the mixed results identified in the current systematic review. Reductions in FA are often reported in DTI studies, and interpreted as disruptions to “white matter integrity” (Jones et al., 2013; O'Donnell and Pasternak, 2015). However, FA reductions simply reflect a general loss of directional water diffusion along a principle direction within a voxel, and can arise from multiple pathophysiological factors. In the chronic stage of TBI, decreases in anisotropy are thought to be predominantly driven by secondary pathological processes, such as demyelination, axonal degeneration, vasogenic oedema, or ongoing inflammatory responses (Assaf and Pasternak, 2008; Beaulieu, 2002). Interestingly, one study included in our review reported significantly higher FA in the internal capsule (Lo et al., 2009), which is somewhat unexpected given the general pattern of FA reductions or normalisation found in other studies assessing chronic TBI. Prior research has suggested that FA increases may reflect cytotoxic oedema; however, this is more often observed in the acute or subacute phases of injury (Lindsey et al., 2023; Mito et al., 2022; Wilde et al., 2008). Another possible explanation of this may be related to selective white matter degeneration in crossing fibre regions, where selective axonal loss may result in an apparent increase in FA in the presence of the relatively preserved crossing fibres (Mito et al., 2018). Nevertheless, the pooled results of our meta-analysis showed that individuals with chronic TBI can be differentiated from controls by overall reductions in FA and increases in MD within white matter tracts. Among the white matter ROIs examined in the current review, the CC was the most commonly reported region and exhibited the most consistent results (i.e., reduced FA and increased MD). This finding aligns with previous systematic reviews regardless of injury severity and chronicity (Aoki et al., 2012; Charney et al., 2026; Hulkower et al., 2013; Wallace et al., 2018).. The CC is the largest white matter structure that connects between the two cerebral hemispheres, and its central midline location likely renders it more vulnerable to shear and strain forces during direct impact (Johnson et al., 2013; McAllister et al., 2012). Recent experimental models of TBI provided further evidence of a direct correlation between biomechanical strain and dMRI measures of axonal injury in the CC (Donat et al., 2021). Moreover, previous studies have suggested region-specific differences in injury susceptibility within the CC, where the splenium and genu are thought to be more affected than the body (Champagne et al., 2019; Rutgers et al., 2008). An earlier meta-analysis of all-stage TBI reported similar findings that FA reductions were most pronounced in the splenium, followed by the genu and then the body, in both mild and more severe injuries (Wallace et al., 2018). However, in our study, we found greater FA reduction in the body than in the genu in mild TBI, and the greatest FA reduction in the body compared with the extremities of the CC in moderate-to-severe injuries. The factors contributing to these divergent findings between prior analyses and ours are unclear, but they might be related to differences in study inclusion and exclusion criteria. The aforementioned meta-analysis did not explicitly exclude sports-related injuries and had no restriction on time post-injury (Wallace et al., 2018), whereas our included samples were limited to individuals with chronic TBI from the general population. Although the CC remains an important white matter tract that is frequently implicated in TBI pathophysiology, FA reductions and MD increases were also found in many other ROIs. A similar pattern of widespread changes in diffusion measures across white matter tracts following TBI has been reported previously (Wallace et al., 2018). Our meta-analyses showed that in mild TBI, the centrum semiovale and corticospinal tract exhibited larger FA effect sizes than any subregion of the CC, although only the effect in the centrum semiovale was significant. Likewise, for MD, the middle cerebellar peduncle showed a larger and significant effect size than the CC in mild TBI. Further, in moderate-to-severe injuries, significantly lower FA was also observed in many other fibre tracts, including the superior longitudinal fasciculus, inferior fronto-occipital lobes, external capsule, cingulum, cerebral peduncle, and sagittal stratum. However, it should be noted that the number of studies investigating dMRI measures in these ROIs was very limited. In fact, many studies have taken a priori approaches when conducting ROI analyses and thus tend to focus on regions commonly associated with TBI, while underrepresenting other white matter tracts. Future studies are encouraged to examine a broader range of white matter ROIs and to perform whole-brain level analysis, with consistent reporting of diffusion metric findings across all regions. These approaches may help better capture brain-wise microstructural changes and improve understanding of the spatial patterns of white matter alterations following chronic TBI. Our findings also show considerable variation in effect sizes across studies. Variance due to between-study heterogeneity rather than sampling error was observed for several white matter ROIs, as indicated by the I2 values. The prediction intervals, which reflect the range of true effects, further highlight the existing variance across studies. Potential sources of heterogeneity were identified from meta-regression analyses, including age, sex, and magnetic field strength. Although the remaining variables we selected could not adequately explain the observed heterogeneity, it is worth noting that the study size included in the meta-regression model was relatively small, which might have reduced the possibility to detect such effects. Besides, other clinical-demographic and imaging factors that were not selected as moderators may also influence the effect estimates. For example, the GCS score is considered an important prognostic factor for long-term outcomes in TBI (Teasdale et al., 2014), yet it was not reported in over half of the studies included. Moreover, differences in how white matter ROIs were defined in each study may have contributed to the variation in effect sizes. This is particularly relevant for studies in which ROIs were manually drawn, as this procedure can be subject to the operators' decisions as well as by the size of the ROIs selected. Furthermore, focal lesions may represent a potential confounding factor in studies examining participants with moderate and severe TBI. The presence of lesions could bias results in several ways, such as by affecting registration performance to an image template, causing inaccuracies in ROI delineation, or directly influencing the estimation of quantitative diffusion measurements. Whether focal lesions were used as an exclusion criterion, included in the analysis, or handled properly during image processing could all result in between-study heterogeneity. More specifically, differences in the size, location, and number of focal lesions may also contribute to variability in the results. However, in the studies included in the current review, detailed information on lesion presence and handling strategies was largely underreported (see Supplementary Table S11 for details). To minimise sources of data heterogeneity and enhance comparability of findings across different sites, future studies should aim to control for all key confounding variables during subject recruitment, such as age, sex, injury severity, time since injury, and all relevant clinical factors including comorbidities, presence of focal lesions, and prior history of neurological or psychiatric disorders. Future studies should also consistently report detailed and sufficient information on sample characteristics, dMRI acquisition protocols, and image processing procedures to enable meaningful comparisons and meta-analytic data synthesis. When feasible, future studies are encouraged to provide complete data on the outcome measures and avoid selective reporting. Overall, the present meta-analysis demonstrates diffuse white matter changes in the chronic phase following both mild and moderate-to-severe TBI. Importantly, these effects were observed in the general population, predominantly comprising individuals with a single TBI and with no representation of sports- or combat-related injuries. This underscores the need for ongoing research focusing on the chronic phase of TBI, aimed at better characterising long-term white matter microstructural changes, as well as understanding their associations with neurocognitive functioning and outcomes. Limitations There are a number of limitations to be considered in relation to both the studies included in the review and the review process itself. First, although we included a total of 27 studies in the meta-analysis, the sample size of individual studies was relatively small. This was particularly evident in the estimation of effect sizes for individual white matter ROIs, which may have reduced the statistical power and precision of these estimates. Next, there are concerns about risk of bias in the included studies, mostly due to a lack of control of confounding factors, such as whether individuals with TBI had a history of other neurological disorders or were exposed to medications. A summary of the risk of bias assessment is reported in the Supplementary Material. We note that the implementation of a quality checklist for dMRI studies (Imms et al., 2019; Jiang et al., 2017) could provide additional context regarding the methodological quality of the included studies; however, given that there is not an established tool or gold standard for assessing the quality of dMRI studies, we did not perform such an assessment in the current review. Another limitation to be considered is the potential for publication bias. The assessment of publication bias revealed the presence of small-study effects in moderate-to-severe TBI studies, which might have led to an overestimation of the pooled effects. In the sensitivity analysis restricted to studies with sample sizes >10, the funnel plot again showed asymmetry (Supplementary Fig. S5), suggesting that small-study effects were still present even after excluding the smaller studies. This indicates that the observed bias is unlikely to be solely driven by the inclusion of very small studies, and may instead reflect broader sources of plot asymmetry. Of note, we were unable to perform sensitivity analyses for MD due to insufficient data, which limited direct comparison with FA in terms of robustness of the findings. With regard to the inclusion of studies using ROI analysis, although this approach allowed us to investigate individual white matter tracts, it may have introduced bias because the definition of ROIs can be either operator-dependent or hypothesis-driven. Alternatively, whole-brain voxel-based analyses or tract-based spatial statistics (TBSS) (Smith et al., 2006) may offer the advantage of providing greater coverage of white matter changes across the entire brain. TBSS provides a means to mitigate image misalignment issues in whole-brain DTI analyses, and has been widely implemented in TBI studies (Hulkower et al., 2013). However, we excluded studies reporting TBSS-based ROI findings in our meta-analysis, as the skeletonisation step in TBSS limits ROIs to a small number of voxels, which may increase bias in the reported effect sizes. Furthermore, we focused only on FA and MD as outcome measures, and did not include other DTI-based metrics (e.g., axial and radial diffusivity) in the meta-analysis. While these additional metrics could be informative, we note limitations in their interpretability (Wheeler-Kingshott and Cercignani, 2009). Additionally, in this study, we focused on DTI metric findings themselves in chronic TBI, rather than on their association with long-term clinical or functional outcomes. Future work that explores these associations may be valuable, although it relies on consistent outcome data that are not always available in TBI studies. Despite the widespread use of FA and MD in TBI studies, these metrics are known to be limited in their interpretability due to inherent limitations to the DTI model (Jones et al., 2013). The DTI model assumes that the displacement of water molecules occurs following a Gaussian distribution, which is invalid in biological tissues comprising complex cellular structures (Wu and Cheung, 2010). Consequently, DTI is particularly unreliable for characterising voxels containing fibre bundles with different orientations (i.e., ‘crossing fibres’), which constitute up to 90% of white matter voxels (Jeurissen et al., 2013). At the typical image resolution of dMRI (∼1–2 mm3 isotropic), the tensor metrics may fail to accurately represent the diffusion properties of the underlying tissue microstructure (i.e., axonal structures with diameters on the order of μm). To achieve a better understanding of white matter microstructural changes in chronic TBI, advanced diffusion models are likely to be beneficial. For example, DKI has been developed to characterise non-Gaussian properties of water diffusion (Jensen et al., 2005), and mean kurtosis (MK) can be derived from the model to quantify the average deviation of water diffusion from a Gaussian form (Wu and Cheung, 2010). Among studies included in the current meta-analysis, one employed both DTI and DKI to investigate white matter changes in individuals with TBI more than one year after injury. This study found significantly lower FA and MK in the thalamus and splenium of the CC, and significantly lower MK in the centrum semiovale (Grossman et al., 2012). Moreover, higher MK in the thalamus was significantly associated with better neurocognitive performance, whereas FA was not correlated with neurocognitive functioning. Another study included in the review also examined changes in the mean of the kurtosis tensor (MKT) in addition to FA (Næss-Schmidt et al., 2017), with MKT being a variant of MK derived from DKI (Hansen et al., 2013). Significantly lower MKT was observed in the thalamus after mild TBI, while no significant group differences were found in FA between individuals with chronic TBI and controls. Taken together, these studies have demonstrated that the use of diffusion kurtosis metrics could complement DTI metrics and may lead to improved detection of regional white matter changes following chronic TBI. In addition to DKI, other advanced dMRI techniques include fixel-based analysis (FBA) (Raffelt et al., 2012; Raffelt et al., 2017), which enables characterisation of specific fibre populations even within crossing-fibre regions, and microstructural compartment models (e.g., NODDI) that can provide insight into different tissue microstructures (Zhang et al., 2012). Some of these advanced dMRI approaches have been implemented in the context of TBI, offering novel perspectives into the characterisation of white matter changes. Specifically, FBA studies have demonstrated significant reductions in fibre density and fibre cross-section across multiple tracts following chronic TBI (Clemente et al., 2023; Jia et al., 2026; Wallace et al., 2020a). FBA-derived metrics also enable the identification of differential damage in crossing fibre regions that cannot be resolved by voxel-wise FA, thereby improving specificity to microstructural abnormalities in both acute and chronic stages of TBI (Jia et al., 2026; Mito et al., 2022; Wright et al., 2021). However, in this work, we were unable to identify a sufficient number of studies to perform a statistically robust meta-analysis of diffusion metrics derived from these advanced techniques. Such a review will be warranted once more studies report their findings using advanced dMRI analysis. Conclusion This study provides quantitative evidence of persistent and widespread white matter alterations in the chronic phase of TBI, as detected by dMRI metrics. Higher MD and lower FA values are observed in individuals with chronic TBI from the general population compared with controls, and the effects become more pronounced in moderate-to-severe TBI than in mild injuries. While the CC remains the most extensively studied tract and shows marked white matter changes, other white matter tracts also exhibit structural abnormalities that warrant further investigation. Our findings on the associations between FA effect sizes and clinical-demographic and imaging variables highlight potential sources of heterogeneity between studies, such as age, sex, and magnetic field strength. These factors may partially account for the variability in reported outcomes within the current literature and should be carefully controlled for in future research. CRediT authorship contribution statement Yining Chen: Writing – review & editing, Writing – original draft, Project administration, Methodology, Formal analysis, Data curation, Conceptualization. Jessica Hoffmann: Writing – review & editing, Methodology, Investigation, Data curation, Conceptualization. Warda T. Syeda: Writing – review & editing, Supervision, Methodology, Investigation, Formal analysis, Data curation, Conceptualization. Remika Mito: Writing – review & editing, Supervision, Methodology, Investigation, Formal analysis, Data curation, Conceptualization. Bradford A. Moffat: Writing – review & editing, Supervision, Project administration, Methodology, Conceptualization. Declaration of competing interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.