Introduction

Migraine is a major neurological disorder characterized by moderate or severe headaches with a unilateral, pulsatile quality accompanied by a myriad of symptoms, such as nausea, photophobia, and phonophobia. There are several categorizations for migraine, such as chronic (CM) and episodic migraine (EM), migraine with aura (MWA), and migraine without aura (MWoA), i.e., the presence/absence of sensory disturbances, including flashes of light, blind spots, and hand/face tingling, respectively ("Headache Classification Committee of the International Headache Society (IHS) The International Classification of Headache Disorders, 3rd edition," 2018). According to the global burden of disease study in 2017, headache disorders were the second most prevalent disease and the second-highest contributor to age-standardized global years lost due to disability (YLD). Among headache disorders, migraine was in the first rank based on YLD and in the second rank based on prevalence (Sevenich, 2018). Despite the high prevalence and the well-known clinical features of this disease, the cascade of events triggering initiation and disease progression is far from being understood.

Recently, it has been suggested that alterations in mechanisms involved in cortical excitatory-inhibitory balance may promote hyper-reactivity to pain in individuals genetically susceptible to migraine (Gasparini et al., 2013; Mainero et al., 2011). Because of this increased neuronal excitation, a wave of cortical spreading depression (CSD) turns up and spreads across the cerebral cortex (Charles & Baca, 2013; Granziera et al., 2006). Animal studies showed that neuronal excitation and the subsequent CSD might lead to several alterations, such as activation of trigeminal afferent neurons (Karatas et al., 2013; Moskowitz et al., 1993), increasing brain vascular permeability, and promoting neuroinflammation (Cutrer et al., 2012; Gursoy-Ozdemir et al., 2004). The latter can trigger neuronal firing in the spinal and trigeminal nucleus and in the meningeal nociceptors (Kincses et al., 2019). Thus, the recurrent sensitization of the trigeminovascular system linked with the increased reactivity to stimuli is considered the main trigger of the cascade of events associated with migraine attacks (DaSilva et al., 2007; Welch, 2005).

The trigeminal complex projects to the brainstem, hypothalamus, basal ganglia, thalamic, and cortical regions involved in sensory and cognitive pain input and aura (Dodick, 2018; Noseda et al., 2011). Projection neurons can modulate central signals from the periaqueductal gray matter (PAG), dorsolateral pons, medullary raphe, spinal trigeminal nucleus (SpV), as well as the descending cortical inhibitory complex (Marciszewski et al., 2018). These areas mediate the intensity of sensory stimuli, cerebral blood flow, and nociception of cortical and subcortical neurons (Dodick, 2018; Maniyar et al., 2014) exhibiting different levels of activity in the migraine stages (Dodick, 2018; Moulton et al., 2008). Similarly, the thalamus, a bilateral brain structure projecting out to the cerebral cortex through several tracts, such as the fornix, cingulum, anterior thalamic radiations (ATR) and posterior thalamic radiations (PTR) (Jones, 2002; Zhang et al., 2010), is involved in the pathophysiology of migraine (Li et al., 2011; Yuan et al., 2012).

Despite these significant advancements in the comprehension of the potential neural mechanisms linked with migraine, we still lack effective imaging biomarkers aimed at predicting the disease course or treatment response (Dodick, 2018). In this regard, the analysis of the human connectome, i.e., the human brain organization into highly interconnected regions (Sporns et al., 2005), could help identify new non-invasive cost-effective biomarkers for diagnosis, progression, and as surrogate outcomes for clinical trials (Dodick, 2018; Katsarava et al., 2012). In the last years, diffusion-weighted imaging (DWI) sequences, based on the difference in magnitude of water diffusion, have improved our knowledge of neurological disorders. Diffusion tensor imaging (DTI) comprises a group of techniques computing eigenvalues (λ1, λ2, and λ3) and eigenvectors (ε1, ε2, and ε3) used to define an ellipsoid that represents an isosurface of diffusion probability aimed at understanding the microstructural properties of the brain tissue (Huisman, 2010; O'Donnell & Westin, 2011; Zhang et al., 2020). Four DTI indices are commonly used to quantify the shape of the tensors in each brain voxel. The fractional anisotropy (FA) is the most widely used anisotropy measure, an index of the amount of diffusion asymmetry within a voxel. When λ1 = λ2 = λ3, the diffusion ellipsoid is a sphere indicating a perfect isotropic diffusion (FA = 0). With progressive diffusion anisotropy, the eigenvalues become more unequal, and FA values became higher. A complementary measure to FA is mean diffusivity (MD) computed as the average of the three eigenvalues of the tensor. Finally, AD and RD could be helpful in determining the diffusivity direction, along the main axis (λ1) or perpendicular to it (average of λ2 and λ3). FA is sensitive to axonal integrity, although many factors are linked with FA changes (i.e., cell death, gliosis, demyelination, increase in extracellular or intracellular liquid content, inflammation, and axonal loss). Therefore, FA is not a specific parameter to define the type of changes (Neeb et al., 2015; O'Donnell & Westin, 2011; Zhang et al., 2020) and is usually paired with MD. High MD indicates increased extracellular spaces because of shrinkage or degeneration of axons and dendritic fibers. Thus, MD is higher in cerebrospinal fluid (CSF) compared to GM and WM, as water molecules can move freely (Narr et al., 2009; Tromp & Scalars, 2016). Finally, AD and RD could be used to detect axon myelination or pathology (Zhang et al., 2020). AD is sensitive to axonal degeneration, which is associated with fiber density and axon intrinsic characteristics (Messina et al., 2015; Neeb et al., 2015). Whereas demyelination, abnormal axonal diameter, or density may influence RD (Messina et al., 2015; Tromp & Scalars, 2016).

According to a recent coordinate-based meta-analysis consisting of both volume and surface GM and DTI studies, there is no clear consensus about brain structural alterations in migraine (Masson et al., 2021a, b). However, several DTI studies showed widespread alterations of the diffusivity metrics, suggesting a multifaceted association between migraine and brain structural connection/organization (Kim et al., 2021). Herein, we aim to systematically review DTI studies and comprehensively discuss microstructural changes in migraine. Moreover, we aim to clarify whether these changes are associated with clinical parameters, including attack duration, frequency, disease duration, and different phases of migraine.

Methods

This systematic review was performed in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines (Moher et al., 2009).

Literature search and selection criteria

We performed an online search in PubMed and Scopus databases in January 2022. The search terms included "Diffusion Tensor Imaging OR Diffusion Magnetic Resonance Imaging OR Diffusion-Weighted Imaging OR Fractional Anisotropy OR Diffusivity OR Tractography" AND "Migraine OR Migraine Disorders OR Migraine Headaches OR Migraine with Aura OR Migraine without Aura OR Chronic Migraine OR Episodic Migraine", and the equivalent search terms in each database. Reference lists of the included studies and other relevant studies were also reviewed for eligible studies.

Original studies in English were included if they (1) measured tract-based or region of interest (ROI) diffusion metrics through computational DTI methods and (2) compared microstructural changes in patients with migraine with healthy controls (HCs) or microstructural features between patients with different migraine types (e.g., with or without aura).

We excluded (1) case reports, case series, letters, commentaries, abstracts, review articles, and animal or in vitro studies, (2) studies including patients diagnosed with different neurologic conditions, and (3) interventional studies.

Data selection was performed in concordance with the PRISMA guidelines (Moher et al., 2009). Two authors (RR and MHA) independently assessed the eligibility criteria of the studies. In case of conflicting judgments, a third author’s (MD) opinion was asked.

Data extraction

The extracted data included: (1) demographic features of the samples, including age and sex of patients and HCs, (2) data related to the disease characteristics, including classification of migraine (with or without aura), disease duration, and attack frequency and duration, (3) the characteristics of image acquisition, including field strength and b-value, (4) DTI analysis methodology, (5) the spectrum of data analysis (whole brain or tract-based), (6) key findings, including the alterations of diffusion metrics across brain regions or tracts, and (7) other relevant findings.

Results

Study selection

The PRISMA chart for studies selection is depicted in Fig. 1. A total of 646 articles were identified. After removing duplicate records, title and abstract of 409 studies were screened, leading to the exclusion of 361 studies. Of the remaining 48 studies that entered full-text screening, 35 studies were finally included. Thirteen studies were excluded due to the following reasons: network-based DTI analysis (six studies), histogram analysis in patients with WM lesions (three studies), using of imaging modalities other than DTI (two studies), interventional design (one study), and participants with a different neurological condition (one study). Table 1 summarizes the included studies. Reviewed articles included data from 2220 individuals (574 males) consisting of 1253 individuals diagnosed with migraine (295 male patients) by the International Classification of Headache (ICHD) criteria (second and third edition) (Headache Classification Committee of the International Headache, 2013; "Headache Classification Committee of the International Headache Society (IHS) The International Classification of Headache Disorders, 3rd edition," 2018; Olesen & Steiner, 2004), and 967 HCs.

Fig. 1
figure 1

PRISMA flow diagram for the systematic review of diffusion tensor imaging studies in migraine headache patients

Full size image
Table 1 Diffusion imaging studies included in the review
Full size table

Study characteristics

Data for mean disease duration (years since patients received migraine diagnosis) and attack frequency (the number of migraine headache attacks per month or days) were reported for all articles but Delic et al. (2016), Gomez-Beldarrain et al. (2015), and Kattem Husoy et al. (2019). Mean disease duration ranged from less than three years (Messina et al., 2015) to about 34 years (Schmitz et al., 2008) among the included articles. Several studies reported mean attack duration, defined as the number of hours each attack lasts (Coppola et al., 2014, 2020; Liu et al., 2011; Li et al., 2011; Mettenburg et al., 2012; Yu et al., 2013b; Zhang et al., 2012). Comparing studies which investigated depression and migraine as separate entities and studies focused on concomitant migraine and depression could be helpful in unraveling the distinction between brain diffusivity alterations inherent to these disorders. As decreased AD in WM tracts such as CC, IC and EC was reported in both depressed and non-depressed patients with migraine, axonal loss might be attributable to migraine pathophysiology rather than depressive symptoms, while RD increase and FA reduction, which are presumed markers of demyelination, are observable more frequently in MWOA patients with depressive symptoms (van Velzen et al., 2020; Yu et al., 2013b). Thus, it can be assumed that axonal degeneration and brain atrophy might highlight the adaptive reaction of neurons in response to frequent migraine attacks, while demyelination might represent the main response to depressive symptoms (Li et al., 2011). Future studies should confirm these assumptions.

Correlation between clinical variables and DTI indices in migraine

Some of the studies discussed in the present review assessed the correlation between age and DTI indices with contrasting results. It is worthy of note that participants of the included studies were mainly middle-aged, a period of time when microstructural changes are minimal (Behler et al., 2021). For the aim of understanding age-related microstructural changes in migraine, DTI metrics should be carefully evaluated through longitudinal studies. Similarly, no correlations were reported between the main DTI indices with disease duration, frequency of migraine attacks, and pain intensity. However, preliminary evidence reported a significant positive association between both attack frequency and disease duration with higher MD and lower FA mainly within the CC, thalamic radiations, and CST. These findings, although inconsistent and heterogeneous, might suggest potential deleterious effects of frequent migraine attacks on myelination, which might emerge as a function of the severity of the disease. This might explain the heterogeneity of the findings, although further studies should assess this relationship to unravel the underlying pathophysiological mechanisms influencing brain microstructure by migraine attacks.

Beyond diffusion tensor imaging

Diffusion MRI investigate microstructural structures in vivo in the biologic tissue, detecting early WM microstructural changes. However, to date, DTI indices are non-specific biomarkers for several neurological disorders, including migraine (Alexander et al., 2007). The lack of specificity might depend on some limitations, including subject motion, image resolution, partial volume effect, and crossing WM fibers resulting in potentially biased FA computation (Alexander et al., 2007; Pasternak et al., 2018). Furthermore, DTI technique owns several intrinsic limitations in the detection of GM abnormalities. Indeed, FA can detect water diffusion restriction in anisotropic areas, while GM, consisting of neuronal body, has isotropic properties as diffusion of water molecules is not restricted (Ghazi Sherbaf et al., 2018). These issues might confound the results of migraine studies, masking some potential effects.

Additionally, numerous WM brain structures consist of several complex fiber arrangements (e.g., the CR and the WM adjacent to the cortex), and due to the multiple cross-fibers, DTI indices in these tracts might be suboptimal (Deligianni et al., 2016; Pasternak et al., 2018). Moreover, in neurological disorders, brain microstructure can be affected by the combination of several alterations such as gliosis, inflammation, demyelination, axonal loss and plastic neuronal changes, which are known to affect DTI metrics, making the interpretation of such alterations problematic (Alexander et al., 2007; Pasternak et al., 2018). Moreover, DTI model assumes Gaussian distribution for diffusion within each voxel (Pasternak et al., 2018; Winston, 2015), which is not necessarily the case at the whole brain level (Ghazi Sherbaf et al., 2018). These limitations highlight the necessity to apply novel advanced modalities in migraine, such as Diffusion kurtosis imaging (DKI) and neurite orientation dispersion and density imaging (NODDI). These imaging methods might improve the sensitivity, detecting subtle alterations. DKI is based on non-Gaussian diffusion (Ito et al., 2016), which might improve the detection of diffusion abnormalities in both isotropic (e.g., GM) and anisotropic regions based on the degree of diffusion restriction (Ghazi Sherbaf et al., 2018). DKI consists of three main kurtosis indices, including mean kurtosis (MK), axial kurtosis (AK), and radial kurtosis (RK), representing the structural complexity. Overall, these parameters showed higher sensitivity for detecting crossing fibers compared to DTI and are less prone to the partial volume effect and CSF contamination.

NODDI is a novel model to detect morphology of neurites. Similar to DKI, the model is more suitable for both GM and WM, and less CSF contamination is expected for this modality. Nevertheless, NODDI can be applied to diffusion MRI data typically acquired in clinical setting, which might hasten its extensive application in the clinical research field (Winston, 2015). Other techniques might provide valuable microstructural information in migraine, such as the diffusion ensemble average propagator (EAP), although this methodology requires multi-shell acquisitions (and longer acquisition time), making it less feasible within the clinical practice. This limitation can be overcome by applying novel approaches. Apparent Measures Using Reduced Acquisitions (AMURA), a technique assuming that diffusion anisotropy is approximately independent of the radial direction (Aja-Fernández et al., 2020), could reduce the acquisition time. This approach has been recently applied in migraine patients, showing promising results to detect microstructural changes associated with this disorder (Planchuelo-Gómez et al., 2020). Further studies should evaluate the application of advanced statistical and machine learning techniques aimed at assessing the latent relationships between migraine features and microstructural changes, such as multimodal canonical correlation analysis combined with joint independent component analysis (Planchuelo-Gómez et al., 2021).

Limitations and future direction

Despite its strengths, this systematic review is prone to some limitations. About 20% of the included studies were drawn from similar, or partially overlap**, samples of participants, which might have influenced some results. Moreover, despite our efforts to discuss both significant and non-significant results, there is a tendency towards publication of papers with significant results compared to non-significant ones (i.e., publication bias). Indeed, most of the studies reported significant changes in at least one diffusion parameter, while only nine studies reported completely non-significant results among the investigated DTI parameters. Furthermore, the studies included in this review might be prone to some intrinsic limitations, which should be addressed by future studies. First, the number of patients included in most of these studies was lower than 50, which might limit the generalizability of the conclusion that can be drawn. Additionally, as reported in the present review, imaging time at different phases through migraine cycle might influence the results, due to the fluctuations of DTI parameters in the different phases of migraine. Some possible confounding factors (e.g., positive family history of migraine, severity of pain, presence or absence of aura, types of auras, and anxiety/depression profiles of participants, patient's age and sex, medication intake) might influence DTI findings. Similarly, heterogeneity in the study design, clinical features, and analysis methods might lead to conflicting results, making the interpretation of microstructural alterations in migraine more complex (Forkel et al., 2020). Depression also has an effect on brain microstructure, especially the genu of the CC and the left ALIC (Chen et al., 2016). Therefore, it should be considered for patient selection or classification. Moreover, no study has assessed whether a positive family history of migraine or a genetic predisposition is linked to WM microstructure alterations. Finally, most of the studies were cross-sectional, thus it is impossible to infer a cause-effect relationship between migraine and DTI. Longitudinal studies will be extremely helpful to investigate the development of microstructural abnormalities during disease or treatment.

Conclusion

Despite the great effort to investigate the pathophysiology of migraine, the evidence summarized here suggests that future studies are still necessary to unravel brain diffusion alterations linked with this disorder. Preliminary evidence suggest that microstructural alterations occur during the disease. Reduced microstructural integrity was observed in the thalamus, CC, longitudinal fasciculus, and cingulum in patients with migraine compared to controls. However, the tensor model was unable to find remarkable differences between different migraine subtypes. Notably, changes in DTI indices occur in the interictal phase, which might be interpreted within the habituation deficit theory or as neuronal plasticity mechanisms. Moreover, these results might suggest that frequent stimulation and CSD events lead to release of neurotransmitters and pain generation, and finally, cellular damage, as captured by DTI indices. Indeed, repetitive occurrence of neuronal damage can be associated with disruption of WM microstructure and decreased FA in several brain microstructures. In chronic migraine, the variable equilibrium between neuronal damage, due to repetitive pain stimulations, and plastic neuronal changes occurring as compensatory mechanisms might lead to higher heterogeneous results. DTI assessment and interpretation of structural abnormalities in migraine are still questionable due to the complexity of migraine pathophysiology and DTI limitations. Further longitudinal studies, applying novel advanced modalities are required to fully understating the effects of migraine on brain structural connectivity and its progression.