In our study, we examined the topological alterations in the structural brain connectome of participants with EM and CM. Diminished global efficiency and increased characteristic path length was found in the CM group compared with the HC group, indicating a less efficient structural network and longer information transfer pathways. Both EM and CM groups exhibited significantly reduced local efficiency compared to HCs, with CM showing lower local efficiency than EM. Moreover, the CM group demonstrated significant reductions in local clustering coefficient and nodal local efficiency in frontal and temporal brain regions compared to HCs and EM group. Nodal local efficiency effectively differentiated between CM, EM, and HC groups. Furthermore, the nodal local efficiency of specific brain regions, such as the left opercular part of the inferior frontal gyrus and right middle frontal gyrus, were negatively correlated with attack frequency. These findings align with the pathophysiology, implicating frontal and temporal lobe brain networks in anxiety, depression, and pain modulation in migraine [11].

Previous studies have laid the groundwork for understanding the intricate details of white matter micro-structural imaging in the context of migraine with voxelwise approaches [30]. Yu et al. have found that patients with EM showed significantly lower FA in several brain regions, including the subcortical white matter of frontal lobe, temporal lobe, and parietal lobe [12]. Similarly, another study using tract-based spatial statistics (TBSS) analysis revealed widespread increases in radial diffusivity (RD) and mean diffusivity (MD) values in the CM group compared to HCs [13].. In our study, the NBS analysis revealed distinct reductions in FA connectivity components in both EM and CM groups compared to HCs, showcasing specific nodes and connections implicated in each condition. While both patients with CM and EM exhibited micro-structural damage, as measured by TBSS, graph theory analysis revealed that patients with CM displayed greater significant alterations at the network level. Similarly, examining global topological network features in the context of the chronification of migraine, individuals with CM exhibited significant alterations compared to control participants. In contrast, those with EM showed minimal abnormalities. Based on our findings and existing literature, we propose that the chronification of migraine is not solely attributable to micro-structural alterations in white matter. Instead, it is suggested that a severe disruption of structural connections between brain areas, forming a network, is necessary to induce changes in information integration and organization, leading to migraine chronification.

Recent studies have highlighted the structural and functional connectivity alterations in migraine, revealing critical insights into its pathophysiology. For instance, Michels et al. observed that migraine patients exhibited a more segregated network topology, with CM patients showing greater modularity compared to EM, suggesting maladaptive reorganization in headache-related brain circuits [31, 32]​. Similarly, Dai et al. reported enhanced integration and efficiency in global network properties among EM patients, correlating with clinical measures such as disease duration and headache impact scores [33]. In CM, DeSouza et al. found reduced global and local efficiency alongside increased segregation, with disruptions prominently in the limbic and insular cortices [34]. Structural network alterations extending to pain processing and modulation regions, such as the posterior cingulate and inferior parietal lobule, were further emphasized by Silvestro et al., who introduced a connectopathy model for migraine​ [35]. Li et al. focused on the vulnerability of rich-club regions, which showed increased feeder connection density in migraine patients, enhancing integration within pain-related circuits [36]​. Lastly, Planchuelo-Gómez et al. underscored the coexistence of strengthened subcortical and weakened cortical connections in migraine, providing a nuanced understanding of structural connectivity changes [31]. These findings collectively underscore the importance of investigating global and regional network properties to elucidate the mechanisms underlying migraine chronification and progression. Moreover, recent studies on migraine-related brain networks have provided critical insights into the pathophysiological mechanisms of chronic migraine. Hosp et al. utilized DTI-based global tractography to construct the migraine-related pain network, identifying the insular cortex as a central hub connecting sensory, cognitive, and modulatory pathways, with white matter tract integrity closely linked to self-reported pain levels [37]. Borsook et al. highlighted the importance of subliminal neural dynamics and unconscious brain reorganization preceding chronic pain, suggesting that early interventions could mitigate the transition to chronic pain states [38]. These findings support the present study and further validate the critical role of disrupted structural network efficiency in chronic migraine.

Examining functional connectomes, a magnetoencephalographic study revealed reduced total node strength within pain-related cortical regions (bilateral primary and secondary somatosensory cortices, insula, medial frontal cortex, and anterior cingulate cortex) in CM patients, particularly in the beta band, compared to controls [39]. Notably, negative correlations between attack frequency and node strength were evident in the bilateral anterior cingulate cortex across all migraine patients. In another resting-state functional MRI study, Lee et al. have proposed that patients with CM exhibit enhanced connectivity within the pain matrix compared to those patients with EM [40]. The functional modifications observed in the pain network could contribute to the process of migraine chronification. These functional imaging findings complement our structural network analysis, emphasizing the involvement of frontal, parietal, and temporal lobes in migraine and its chronification.

The global impairment of structural connectivity in individuals with migraine (especially CM) was further supported by the presence of extensively distributed edges exhibiting reduced FA values in CM patients compared to HC participants and patients with EM. The observed decrease in global efficiency and increased characteristic path length in patients with CM suggests disrupted global information transfer and network integration. This aligns with findings in previous DTI studies about migraine, emphasizing the importance of global network properties in understanding headache disorders. The network-based analysis reveals specific alterations in FA values within connected sub-networks. The regions affected in EM and CM groups include brain regions associated with sensory processing and emotions [5]. The involvement of subcortical nuclei (thalamus and basal ganglia) and cortical regions in CM further emphasizes the widespread impact on structural connectivity.

The identification of regions with less local clustering coefficient and nodal local efficiency in both EM and CM groups, particularly in temporal and superior frontal gyrus regions, suggests a disruption in local network organization. The local clustering coefficient is a crucial metric in network science, quantifying the degree of interconnectedness among neighbors of a node in a network [41, 42]. For each node, the local clustering coefficient reflects the extent to which its neighbors form tightly connected clusters, calculated as the ratio of the actual number of edges between the node's neighbors to the maximum possible number of edges. Moreover, in our study, we observed significant group differences in the local clustering coefficient of two brain regions—left middle frontal gyrus, and right dorsolateral part of the superior frontal gyrus (left middle frontal gyrus: HC > EM; left middle frontal gyrus and right dorsolateral part of the superior frontal gyrus: HC > CM; right dorsolateral part of the superior frontal gyrus: EM > CM). These findings suggest that the prefrontal and superior frontal regions, particularly right dorsolateral part of the superior frontal gyrus, exhibit distinct alterations in local clustering patterns during the chronification of migraines. In a functional MRI study, Kong et al. found that in response to high pain stimuli, both the superior frontal gyrus and middle frontal gyrus exhibit significantly increased fMRI signal, suggesting their active involvement in processing and responding to intense pain stimuli [43]. Moreover, Mayr et al. found that in CM, the superior frontal gyrus exhibits reduced activity with increasing pain, indicating its involvement in the altered neural responses associated with CM conditions [44].

Nodal local efficiency is another pivotal metric in network neuroscience, offering insights into the efficiency of information transfer within specific brain regions [45, 46]. Moreover, in our study, we observed significant group differences in the nodal local efficiency of six brain regions—left gyrus rectus, right dorsolateral part of the superior frontal gyrus, right middle frontal gyrus, left opercular part of the inferior frontal gyrus, left hippocampus, and left inferior temporal gyrus (left gyrus rectus, right dorsolateral part of the superior frontal gyrus: HC > EM; left gyrus rectus, right dorsolateral part of the superior frontal gyrus, right middle frontal gyrus, left opercular part of the inferior frontal gyrus, left hippocampus, and left inferior temporal gyrus: HC > CM; right middle frontal gyrus, left opercular part of the inferior frontal gyrus, left hippocampus, and left inferior temporal gyrus: EM > CM). Previous studies have already demonstrated the role of right middle frontal gyrus, left opercular part of the inferior frontal gyrus, left hippocampus, and left inferior temporal gyrus in pain and emotions processing [47,48,49,50,51]. We speculated that the lower efficient information processing of pain circuit within the frontal and temporal lobe may be a biomarker of migraine chronification. This conclusion is consistent with the ROC analysis in which, mean nodal local efficiency about significant group difference exhibited good performance to differentiate among EM and CM participants.

Moreover, the observed negative correlation between local efficiency and MIDAS scores among migraine patients suggests a meaningful connection between changes in local network metrics and the clinical severity of migraine. This association underscores the profound impact of network disruptions on the extent of migraine-related disability, providing deeper insights into the intricate relationship between brain network alterations and the diseases progression of migraine. We also found that the nodal local efficiency of left opercular part of inferior frontal gyrus and right middle frontal gyrus were negatively correlated with the headache attack frequency, which means the disrupted network efficiency of the two regions are significantly associated with migraine chronification.

This study has several strengths and novel contributions. Unlike previous studies that focus on either EM or CM, it comprehensively compares CM, EM, and HCs, revealing distinct structural connectivity disruptions linked to migraine chronification. By integrating multiple graph metrics, such as global/local efficiency and clustering coefficient, this work highlights topological changes, particularly in the frontal and temporal regions, as specific markers of chronification​. Moreover, the associations between these disrupted metrics and clinical features, such as MIDAS scores and attack frequency, enhance its clinical relevance. However, it is important to acknowledge that our study, while providing valuable insights, is not without its limitations. First, this study is an observational study with a cross-sectional design. Cross-sectional studies capture data at a single time point, providing only a snapshot of the disease state. This approach does not account for the temporal dynamics of disease progression or the potential bidirectional relationships between observed network changes and clinical outcomes. Suggest the need for longitudinal studies to identify early imaging markers for disease prediction. Second, the study did not control for the phase of migraine (ictal or interictal) during MRI scanning, which could potentially influence the results. Third, future research should also explore multimodal imaging approaches that combine structural and functional MRI to gain a more comprehensive understanding of the structural and functional changes in migraine [52]. Fourth, while our study primarily focused on cerebral regions, the cerebellum also plays an important role in pain modulation in migraine [53]. This could be considered a limitation, and future research should investigate the involvement of the cerebellum in chronic migraine. Fifth, the choice of using DTI for tractography in this study may not fully capture complex fiber configurations, particularly in regions with crossing fibers. More advanced fiber modeling techniques, such as constrained spherical deconvolution (CSD) or the ball-and-stick model, could provide a more accurate depiction of fiber orientation distributions [54]. Future studies should employ advanced tractography methods to enhance the reliability of connectomic analyses.