Abnormal imaging features of episodic cluster headache and episodic migraine: a multimodal 7T MRI study

Abstract

Background

The objective of this study was to identify functional and structural alterations that differentiate cluster headache (CH) from migraine and address their aberrant imaging features to distinguish from healthy controls.

Methods

Clinical and MRI data were obtained from 87 CH patients, 166 migraineurs (26 with aura and 140 without aura) and 115 healthy controls using a 7T MR system. Between-group differences in the resting-state functional MRI, DTI and T1-weighted structural images were analyzed. Furthermore, the correlation between imaging features and clinical characteristics were explored.

Results

Independent component analysis showed that, compared to controls, CH patients presented decreased fluctuation within auditory network, whereas migraineurs presented decreased fluctuation within default mode network and lateral visual network, and increased fluctuation within salience network (FDR corrected p < 0.05). In comparison with controls, both CH patients and migraineurs had increased connectivity between medial visual network and lateral visual network, and CH patients also had increased connectivity between visual network and somatomotor network (MAFDR corrected p < 0.05). Consistent with inter-network findings, functional connectivity analysis based on anatomical atlas revealed that both CH patients and migraineurs had increased connectivity between brain areas mediating sensory-motor and perceptual information processing. Compared to migraineurs, CH patients had increased functional connectivity among limbic brain regions (FDR corrected p < 0.05). Structural connectivity analysis found that CH patients had increased fiber number between right lentiform nucleus and right cerebellum, and generally decreased fiber number among cerebellar regions (FDR corrected p < 0.05). The vertex-wise analysis showed that CH patients presented reduced cortical thickness in clusters located in the left pericalcarine and right lateraloccipital, compared to both controls and migraineurs (Monte-Carlo cluster corrected p < 0.05). In CH patients, attack duration was negatively correlated with the functional connectivity between medial visual network and lateral visual network, as well as functional connectivity between right calcarine gyrus and right fusiform gyrus. In headache patients, cortical thickness in the left pericalcarine was positively correlated with attack duration and negatively correlated with NRS score (the absolute value of r > 0.2 and p < 0.05).

Conclusions

Generally increased functional connectivity, including between functional networks and between anatomical brain regions, is shared by CH patients and migraineurs. The increased functional connectivity among limbic brain regions and reduced cortical thickness in associated areas may differentiate CH from migraine, potentially reflecting their distinct nociceptive mechanisms.

Supplementary Information

The online version contains supplementary material available at 10.1186/s10194-026-02305-2.

Introduction

Neurological disorders, including migraine and cluster headache (CH), are the leading cause of years lived with disability worldwide [1]. Migraine is a highly prevalent neurological disorder and the third leading cause of disability [1–3]. Migraine is characterized by moderate or severe pulsating headache lasting 4–72 h, accompanied by nausea, photophobia and phonophobia, and aggravated by routine physical activity [4, 5]. CH is characterized by severe unilateral pain lasting 15–180 min with a frequency up to eight per day, predominately in the orbital and supra-orbital region, accompanied by autonomic symptoms ipsilateral to the headache and/or restlessness [6, 7]. The CH attack is considered as one of the most painful experiences, even surpassing childbirth. CH remission occurs on average after a disease duration of 23 years and CH population has a distressing high rate of suicidal ideation [8, 9]. Episodic CH attacks occur in periods lasting from one week to one year (so called “in bout” phase), separated by pain-free periods lasting at least three months (so called “out of bout” phase).

About two thirds of CH patients experience photophobia or phonophobia, and half of CH patients report nausea during attacks. Besides, typical migraine auras are found in some CH patients before or during attacks [10]. On the other hand, about 40% of migraineurs develop unilateral cranial autonomic symptoms during attacks [11]. Thus, duration and particularly restlessness are usually used to distinguish between CH and migraine. But the prevalence of restlessness varies from 48 to 93% in CH population across different countries [12]. Therefore, CH and migraine are often misdiagnosed or delayed diagnosed, resulting in mismanagement or delayed effective treatment among these patients [6, 13]. Our previous large-group study revealed that the first-time diagnostic accuracy rates for migraine and CH were about 27.3% and 23.2% respectively [14].

Findings from MRI imaging can be used to support or supplement diagnosis in patients, especially for those who are difficult to classify by their clinical symptoms alone. Moreover, the pathophysiological mechanisms of these primary headaches are elusive and largely unknown. Some available data are diverse and their findings are often strikingly conflicting [15]. In an attempt to clarify these inconsistencies and provide additional insights, large-scale studies with large sample size and ultra-high-resolution images are warranted. An important consideration for MRI studies should involve data-driven analyses to explore mechanisms underlying these primary headaches and assess differences among different disorders with overlapping clinical features.

Methods

Participants

The study was approved by the Ethics Committee of Chinese PLA General Hospital in accordance with the Helsinki Declaration (S2023-459-01). Written informed consent was obtained from each participant prior to the study. Patients were recruited consecutively from the population attending the International Headache Center of the Chinese PLA General Hospital. Healthy controls were recruited from the general population through advertising. Between May 2023 and April 2025, 97 CH patients, 172 migraineurs and 118 healthy controls were initially recruited for this study. For migraine, only patients with migraine with aura and without aura in interictal phase were enrolled, and patients with chronic migraine were excluded. Episodic CH patients were in the “in bout” phase outside the attacks. Inclusion criteria: (1) definite diagnosis of episodic cluster headache or migraine according to the third edition of international classification of headache disorders (ICHD-3) [16] for headache patients; (2) right-handed; (3) no headache at the time of MRI scan; (4) no medication, drugs, alcohol, smoking, caffeine-containing products or any other substance affecting the central nervous system at least 24 h prior to the MRI scan. Exclusion criteria: (1) fulfilling criteria for more than one ICHD-3 diagnosis; (2) fulfilling criteria for chronic headache; (3) any other neurological, psychiatric or other systemic illness; (4) use of prophylactic headache medication in the past month; (5) brain lesions; (6) MRI contraindications. Besides, healthy controls should have no personal or family history of headache or other chronic pain. Based on the inclusion and exclusion criteria, 10 CH patients, 6 migraineurs and 3 controls were excluded, due to probable diagnosis for not fulfilling all criteria, concurrent diagnosis of both CH and migraine, fulfilling criteria for other ICHD-3 diagnosis, use of headache treatment, concurrent psychiatric conditions or other systemic illness (Fig. 1). Before the MRI scan, the scores of patient health questionnaire-9 (PHQ-9), generalized anxiety disorder-7 (GAD-7) and Pittsburgh sleep quality index (PSQI) were obtained.

Fig. 1.

Schematic diagram of participants’ data enrollment. Abbreviations: CH, cluster headache; ICHD-3, the third edition of international classification of headache disorders; fMRI, functional MRI; ICA, independent component analysis; ROI, region of interest; FC, functional connectivity; DTI, diffusion tensor imaging; FN, fiber number

MRI data acquisition

MRI data of all participants were obtained using the same 7.0-Tesla MR system (MAGNETOM Terra, Siemens Healthineers, Erlangen, Germany). Patients and controls were instructed to keep their eyes closed, not to think of anything in particular, and not fall asleep during the scans. Earplugs were used to reduce noise, and tight but comfortable foam padding was used to minimize head motion. Ultra-high-resolution 3-dimensiomal T1-weighted structural images were acquired using the magnetization prepared 2 rapid acquisition gradient echoes (MP2RAGE) sequence with the following parameters: voxel size = 0.7 × 0.7 × 0.8mm³, repetition time (TR) = 6000ms, echo time (TE) = 2.21ms, inversion time (TI) 1 = 800ms, TI 2 = 2700ms, flip angle (FA) 1 = 4º, FA 2 = 5º, field of view (FOV) = 215 mm×224 mm, matrix size = 288 × 300, slice thickness = 0.75 mm, sagittal slices = 208, acquisition time = 11 min and 54s. Resting-state blood-oxygen-level-dependent (BOLD) functional MRI (fMRI) images were acquired using echo-planar imaging (EPI) sequence with the following parameters: TR = 2000ms, TE = 24ms, FA = 90º, FOV = 216 mm×216 mm, matrix size = 120 × 120, slice thickness = 1.8 mm, slice gap = 0 mm, interleaved slices = 80, volumes = 240, acquisition time = 8 min and 21s. Diffusion tensor imaging (DTI) employed the following parameters: TR = 5120ms, TE1 = 59ms, TE2 = 94ms, FOV = 224 mm×224 mm, matrix size = 140 × 140, slice thickness = 1.6 mm, slice gap = 0 mm, interleaved slices = 80, diffusion values = 0/1000 s/mm², acquisition time = 7 min and 21s.

BOLD MRI data preprocessing

Resting-state BOLD fMRI data were preprocessed using resting-state fMRI data analysis toolkit plus (RESTplus, version 1.31), which is a plug-in software based on statistical parametric mapping (SPM12) in MATLAB platform (version R2022b). The procedure consisted of the following steps: (1) removal of the first 10 volumes for signal equilibrium and participants’ adaptation; (2) slice timing across the remaining 230 volumes; (3) realignment to correct head motion between volumes, and discarding of BOLD data outside the defined motion thresholds (translation > 2 mm or rotational parameters > 2º in any direction) (Fig. 1); (4) coregistration with their corresponding T1-weighted structural images and normalization to the Montreal Neurological Institute (MNI) template using the diffeomorphic anatomical registration through the exponentiated Lie algebra (DARTEL) technique.

Independent component analysis

For independent component analysis (ICA), after first images removing, slice timing, realignment, coregistration and normalization, the preprocessed fMRI data were spatially smoothed with a Gaussian kernel of 4 mm full-width half maximum. ICA was conducted with the group ICA of fMRI toolbox (GIFT, version 4.0b). The number of independent components (ICs) (N = 43) was estimated automated by the software. Firstly, group spatial ICA estimation on the concatenated data from all participants was performed and the data were decomposed into 43 ICs using the infomax algorithm. To ensure the reliability of estimation, the algorithm was repeated several times in ICASSO and the most central run was selected for further analysis. Secondly, spatial maps and time courses were back reconstructed for each participant.

Then, we selected ICs that had peak signal change in gray matter, showed low spatial overlap with known artifacts (vascular, ventricular, motion and susceptibility) and exhibited primarily spectral power at low frequency. This selection procedure resulted in 12 functional networks out of the 43 ICs. Finally, the spatial patterns of the selected components were sorted to match with reported functional network templates [17–19], and further visually inspected to confirm their relevance [20]. These procedures obtained 12 functional networks (Fig. 2A): default mode network (DMN), auditory network (AN), left and right frontoparietal networks (LFPN and RFPN), dorsal and ventral attention networks (DAN and VAN), dorsal and ventral somatomotor networks (DSMN and VSMN), lateral and medial visual networks (LVN and MVN), salience network (SN) and executive control network (ECN).

Fig. 2.

Significant intra-network signal changes of independent components between different groups. (A) Spatial maps of 12 selected independent components. (B) Spatial pattern and average signal change in auditory network of healthy controls and CH patients. (C) Spatial pattern and average signal change in default mode network of controls and migraineurs. (D) Spatial pattern and average signal change in lateral visual network of controls and migraineurs. (E) Spatial pattern and average signal change in salience network of controls and migraineurs. Abbreviations: DMN, default mode network; AN, auditory network; LFPN, left frontoparietal network; RFPN, right frontoparietal network; DAN, dorsal attention network; VAN, ventral attention network; DSMN, dorsal somatomotor network; VSMN, ventral somatomotor network; LVN, lateral visual network; MVN, medial visual network; SN, salience network; ECN, executive control network. L, left. *, FDR corrected p < 0.05

Intra-network analysis of ICs

For assessment of intra-network activation, all participants’ spatial maps for each functional network were entered into SPM12 random-effect analysis by employing one-sample t-test to assess the spatial extent of activation (p < 0.05 corrected using family-wise error [FWE] and cluster extent threshold of 10 voxels), and the obtained spatial extent was used as a mask for subsequent between-group comparison. Then, the average signal change percentage of each cluster was compared between different groups by employing a general linear model, while controlling for sex and age as nuisance covariates (p < 0.05 corrected using false discovery rate [FDR] and cluster defining threshold of p = 0.001).

Inter-network analysis of ICs

For inter-network component connectivity calculation, postprocessing steps were conducted on the time courses of these 12 networks: (1) detrending based on the level of 3; (2) despiking using 3dDespike; (3) filtering with a frequency cutoff of 0.15 Hz. Nest, the Pearson correlation coefficients between pairs of time courses of these networks were estimated, resulting in a 12 × 12 matrix for each participant. To improve normal distribution of correlations, Fisher’s z-transformation was performed. Finally, correlations were compared between groups with a significance level of corrected p < 0.05 (MAFDR corrected), controlling for sex and age as nuisance covariates.

Anatomical functional connectivity construction and analysis

To more precisely specify the functional connectivity (FC) between brain regions, after first images removing, slice timing, realignment, coregistration and normalization, the preprocessed fMRI data were further processed using RESTplus as follows: (1) removing the linear trend; (2) regressing Friston 24 parameters of head motion as well as signals from white matter and cerebrospinal fluid; (3) bandpass filtering with the frequency of 0.01–0.08 Hz to reduce the influence of noise. Unlike ICA, in order to avoid artificial local correlations, no spatial smoothing was conducted herein. To obtain fewer negative connectivity, global brain signals were not regressed.

Region of interest (ROI) based FC construction was conducted using RESTplus by employing automated anatomical labelling atlas (AAL116: 90 nodes in both cerebral hemispheres and 26 nodes in the cerebellum) plus 8 selected subcortical areas that were presumed to be important in headache pathophysiology. Regarding the 8 subcortical areas, we used 6 mm sphere around the peak MNI coordinates (x = ± 6, y = -6, z = -10) for the hypothalamus, 3 mm sphere around the peak MNI coordinates (x = ± 6, y = -36, z = -27) for the dorsal pons, 3 mm sphere around the peak MNI coordinates (x = ± 3, y = -36, z = -45) for the spinal trigeminal nucleus (STN), and 3 mm sphere around the peak MNI coordinates (x = ± 6, y = -30, z = -9) for the periaqueductal gray (PAG) [21]. Then, the time series were extracted from each of these nodes and a 124 × 124 symmetric matrix was obtained by computing Pearson correlation coefficients. Nest, Fisher’s z-transformation was performed to improve the normal distribution of FC values.

The general linear model was performed using graph theoretical network analysis toolbox (GRETNA) to compare FC between different groups (CH vs. control, migraine vs. control, CH vs. migraine). Statistical significance was set at corrected p < 0.05 (FDR corrected), controlling for sex and age as nuisance covariates.

Seed-based voxel-wise FC analysis

Ten ROIs, including bilateral thalamus, hypothalamus, dorsal pons, STN and PAG, were selected as the seed region for correlation analysis. The correlation between the selected seed regions and all remaining brain areas were analyzed using RESTplus software, and Fisher’s z-transformation was performed. Then SPM12 random-effect analysis was employed to assess between-group differences while controlling for sex and age as nuisance covariates (p < 0.05 corrected using FWE).

Structural connectivity construction and analysis

After head motion eddy correction and brain mask extraction using b0 images, DTI data was processed by employing DSI Studio software and structural connectivity matrix based on AAL116 atlas was obtained by deterministic fiber tracking. The minimum fiber length was set to 20 mm [22, 23], and fiber count was set to 1,000,000, which were generated from approximately over 2.2 million seed points. The default settings were used for other tractography parameters as follows: step size = 0.8 mm (automatically calculated as half of the isotropic voxel dimension of 1.6 mm), maximum fiber length = 200 mm, angular threshold = 45°, and fractional anisotropy threshold = 0.08. Then a group mask with a threshold of 0.8 was applied and the fiber number (FN) matrix for each subject was obtained. Finally, a general linear model was performed using GRETNA to compare FN between different groups (CH vs. control, migraine vs. control, CH vs. migraine). Statistical significance was set at corrected p < 0.05 (FDR corrected), controlling for sex and age as nuisance covariates.

Functional-structural connectivity coupling analysis

The nonzero FN edges were selected and rescaled to Gaussian distribution. Their corresponding FC values were extracted from the FC matrix and global brain signals were regressed during the processing steps for coupling analysis. Then the Pearson’s correlation coefficients between the FC and FN matrix were calculated to represent the coupling strength for each participant. The comparison of FC-FN coupling between different groups was conducted by employing GRETNA. Statistical significance was set at corrected p < 0.05 (FDR corrected), controlling for sex and age as nuisance covariates.

Cortical thickness analysis

Cortical model reconstructions were obtained by employing FreeSurfer 7.4.1 to apply on ultra-high-resolution T1-weighted structural images. The segmentation outputs were also visually inspected to check for algorithm accuracy and no manual correction was needed. No subject was excluded due to segmentation errors. The preprocessed data were smoothed with a Gaussian kernel of 10 mm full-width half maximum. Then, the cortical thickness was compared between different groups by employing the general linear model. Statistical significance was set at corrected p < 0.05 (Monte-Carlo simulations with cluster corrected and vertex-wise threshold of p = 0.001), controlling for sex and age as nuisance covariates. Significant clusters were anatomically labelled according to the Desikan-Killiany Atlas.

Correlations between imaging features and clinical characteristics

To explore the possible correlation between aberrant imaging features of patients and clinical characteristics (disease course, attack per year, NRS score, attack duration, PHQ-9 score, GAD-7 score and PSQI score), the Pearson partial correlation analyses were performed using SPSS Statistics (version 29.0, IBM corporation, Armonk, NY, USA), controlling for sex and age as nuisance covariates. The significance level was set at the absolute value of r > 0.2 and p < 0.05.

Results

Demographic and clinical characteristics

Based on the initial inclusion and exclusion criteria, 87 CH patients, 166 migraineurs and 115 healthy controls were prospectively recruited. Further, 8 migraineurs and 3 controls were excluded from fMRI data analysis due to their BOLD data outside the defined motion thresholds (translation > 2 mm or rotational parameters > 2º in any direction). Finally, 87 CH patients, 158 migraineurs (23 with aura and 135 without aura) and 112 controls were included in the fMRI data analysis (Fig. 1). The demographic and clinical characteristics of the participants were presented in Table 1. Age, course of disease, attack frequency per year, PHQ-9, GAD-7, and PSQI scores did not differ between CH patients and migraineurs (p > 0.05). Although there was a tendency of slightly increased PHQ-9, GAD-7, and PSQI scores in CH patients compared to migraineurs, these differences were not significant. Moreover, the distribution of PHQ-9, GAD-7, and PSQI scores showed that most participants fell into the non-clinical or mild range, indicating their limited impact as covariates. As expected, CH patients were predominantly males while most of migraineurs were females, due to their different sex prevalence [24, 25]. CH patients experienced strictly unilateral pain and their NRS scores were significantly higher than migraineurs (p < 0.001). Moreover, the duration of CH attack was generally shorter than migraine (p < 0.001).

Table 1.

Demographic and clinical characteristics of patients and controls

Characteristic	CH	Migraine	Control	CH vs. Control, p	Migraine vs. Control, p	CH vs. Migraine, p
Cases, n	87	158	112
Sex, n (%)				< 0.001 a	0.039 a	< 0.001 a
Male	69 (79.3)	30 (19.0)	38 (33.9)
Female	18 (20.7)	128 (81.0)	74 (66.1)
Age, y #	32 (27–38)	36 (28–42)	31 (26–50)	0.974 a	0.613 a	0.077 a
Disease course, y #	10 (6–15)	10 (7–18)				0.304 b
Attack per year, n #	38 (18–65)	36 (24–72)				0.221 b
NRS score #	9 (8–10)	7.5 (6-8.5)				< 0.001 b
Attack duration, h #	1.5 (1-2.2)	12 (6.5–24)				< 0.001 b
Unilateral pain, n (%)	87 (100)	103 (65.2)				< 0.001 b
left	31 (35.6)	22 (13.9)
right	47 (54.0)	36 (22.8)
left or right	9 (10.3)	45 (28.5)
PHQ-9 #	5 (1–9)	5 (3–7)				0.737 b
GAD-7 #	4 (0–8)	3 (1–6)				0.556 b
PSQI #	6 (4–8)	6 (4–8)				0.366 b

Abbreviations: CH, cluster headache; NRS, numerical rating scale; PHQ-9, patient health questionnaire-9; GAD-7, generalized anxiety disorder-7; PSQI, Pittsburgh sleep quality index

^#, values are presented as median and 25th-75th percentile range. ^a, independent-samples Kruskal-Wallis test is used and significance values are adjusted by the Bonferroni correction for multiple tests. ^b, independent-samples Mann-Whitney U test is used and asymptotic significance is displayed

Intra-network comparisons of ICs between different groups

The spatial distribution of the selected ICs did not differ between different groups. Compared to healthy controls, decreased fluctuation in AN was found in CH patients (corrected p = 0.021, Hedges’ correction effect size= -0.537 [95%CI = -0.820~-0.252]), and the MNI coordinate with the max T absolute value was [-46, -14, 0] (Fig. 2B). Compared to healthy controls, decreased fluctuation in DMN was found in migraineurs (corrected p = 0.023, Hedges’ correction effect size = -0.492 [95%CI = -0.736~-0.246]), and the MNI coordinate with the max T absolute value was [6, -56, 32] (Fig. 2C). Compared to healthy controls, decreased fluctuation in LVN was found in migraineurs (corrected p = 0.035, Hedges’ correction effect size = -0.489 [95%CI = -0.734~-0.244]), and the MNI coordinate with the max T absolute value was [10, -76, 0] (Fig. 2D). Compared to healthy controls, increased fluctuation in SN was found in migraineurs (corrected p = 0.020, Hedges’ correction effect size = 0.546 [95%CI = 0.300~0.792]), and the MNI coordinate with the max T value was [46, -12, 22] (Fig. 2E). No significant difference was found within other ICs between different groups in the average percentage signal change (corrected p > 0.05).

Inter-network connectivity between ICs

Inter-network analysis showed that, compared to healthy controls, CH patients had increased connectivity between MVN and LVN (t = 4.61, p < 0.001), between MVN and VSMN (t = 3.44, p < 0.001), between MVN and DAN (t = 2.94, p = 0.004), between MVN and DSMN (t = 2.40, p = 0.017), between LVN and DSMN (t = 2.83, p = 0.005), between LVN and LFPN (t = 2.62, p = 0.009), between LVN and VSMN (t = 2.61, p = 0.009), between LVN and DAN (t = 2.57, p = 0.011), and between DAN and DSMN (t = 2.65, p = 0.009) (corrected p < 0.05) (Fig. 3A). In comparison with healthy controls, migraineurs had increased connectivity between MVN and LVN (t = 3.37, p < 0.001) (corrected p < 0.05) (Fig. 3B), similar to the finding between CH patients and controls. No significant difference was found between CH patients and migraineurs in inter-network comparison analysis (corrected p > 0.05).

Fig. 3.

Inter-network comparisons of independent components between different groups. (A) Inter-network connectivity alterations in CH patients, compared to healthy controls (MAFDR corrected p < 0.05). Hot colors represent increased connectivity. (B) Inter-network connectivity alterations in migraineurs, compared to healthy controls (MAFDR corrected p < 0.05). (C) Correlation between altered inter-network connectivity and clinical characteristics in CH patients (corrected p < 0.05). Blue colors represent negative correlation. Abbreviations: DMN, default mode network; AN, auditory network; LFPN, left frontoparietal network; RFPN, right frontoparietal network; DAN, dorsal attention network; VAN, ventral attention network; DSMN, dorsal somatomotor network; VSMN, ventral somatomotor network; LVN, lateral visual network; MVN, medial visual network; SN, salience network; ECN, executive control network

In CH patients, a negative correlation was observed between the MVN-LVN connectivity and attack duration (r = -0.264, p = 0.015), between the DSMN-LVN connectivity and GAD-7 score (r = -0.372, p = 0.001), between the DSMN-MVN connectivity and GAD-7 score (r = -0.332, p = 0.004), and between the VSMN-MVN connectivity and GAD-7 score (r = -0.333, p = 0.004) (Fig. 3C).

Group differences in anatomical FC values

FC analysis based on AAL116 atlas plus 8 selected subcortical areas showed that, in comparison with healthy controls, CH patients had increased FC between right supplementary motor area (SMA.R) and bilateral postcentral gyrus (PoCG), between SMA.R and bilateral precentral gyrus (PreCG), between PreCG.R and bilateral paracentral lobule (PCL), between PreCG.R and PoCG.R, between PoCG.R and right calcarine fissure surrounding cortex (CAL.R), between PoCG.R and right cuneus, between CAL.R and right fusiform gyrus (FFG.R), between PCL.L and PCL.R, between left cerebellum 4&5 and bilateral precuneus, and between left cerebellum 3 (CRBL3.L) and vermis 4&5 (Vermis45) (FDR corrected p < 0.05) (Fig. 4A). On the other hand, compared to controls, CH patients had decreased FC between PAG.L and PAG.R (FDR corrected p < 0.05).

Fig. 4.

Comparisons of functional connectivity (FC) based on anatomical atlas plus selected areas between different groups. (A) Anatomical FC alterations in CH patients, compared to healthy controls (FDR corrected p < 0.05). Hot colors represent increased connectivity while cool colors represent decreased connectivity. (B) Anatomical FC alterations in CH patients, compared to migraineurs (FDR corrected p < 0.05). (C) Correlation between altered anatomical FC and clinical characteristics in CH patients (corrected p < 0.05). Blue colors represent negative correlation while red colors represent positive correlation. Abbreviations: PreCG.L, left precentral gyrus; PreCG.R, right precentral gyrus; IFGtriang.R, triangular part of right inferior frontal gyrus; ROL.R, right Rolandic operculum; SMA.R, right supplementary motor area; INS.R, right insula; DCG.L, left median cingulate and paracingulate gyri; DCG.R, right median cingulate and paracingulate gyri; PHG.L, left parahippocampal gyrus; PHG.R, right parahippocampal gyrus; CAL.R, right calcarine fissure and surrounding cortex; CUN.R, right cuneus; IOG.L, left inferior occipital gyrus; IOG.R, right inferior occipital gyrus; FFG.R, right fusiform gyrus; PoCG.L, left postcentral gyrus; PoCG.R, right postcentral gyrus; PCUN.L, left precuneus; PCUN.R, right precuneus; PCL.L, left paracentral lobule; PCL.R, right paracentral lobule; CRBL3.L, left cerebellum 3; CRBL45.L, left cerebellum 4 and 5; Vermis45, vermis 4 and 5; PAG.L, left periaqueductal gray; PAG.R, right periaqueductal gray. The full names of all brain regions were shown in Supplementary TableS1

Although no significant difference was found between migraineurs and healthy controls in anatomical FC values using the threshold of FDR corrected p = 0.05, migraineurs showed a tendency of increased FC between SMA.R and PoCG.R, and between CAL.R and FFG.L (Edge p = 0.001, uncorrected) (Supplementary Fig.S1), similar to the manifestation of CH patients. Likewise, migraineurs had decreased FC between PAG.L and PAG.R, compared to controls (Edge p = 0.001, uncorrected).

Compared to migraineurs, CH patients had increased FC between left parahippocampal gyrus (PHG.L) and right insula, between PHG.L and bilateral median cingulate gyrus, between PHG.R and right Rolandic operculum, between triangular part of right inferior frontal gyrus and bilateral inferior occipital gyrus, between PCL.L and PCL.R (FDR corrected p < 0.05) (Fig. 4B).

In CH patients, a negative correlation was observed between the PAG.L-PAG.R FC and disease course (r = -0.345, p = 0.001), and between the CAL.R-FFG.R FC and attack duration (r = -0.331, p = 0.002). In CH patients, a positive correlation was observed between the CRBL3.L-Vermis45 FC and attack duration (r = 0.281, p = 0.009) (Fig. 4C).

In addition, the seed-based voxel-wise FC analysis showed that, compared to healthy controls, CH patients had increased fluctuation in right inferior parietal lobule when assessing FC with right pontine. By the way, consistently, compared to controls, anatomical FC analysis based on AAL116 atlas plus subcortical areas also showed that CH patients had increased FC between right pontine and right inferior parietal gyrus (p < 0.001), by employing network-based statistic (NBS) approach (Edge p = 0.05, NBS corrected, component p < 0.05, iteration = 5000). On the other hand, in comparison with controls, CH patients had decreased fluctuation in left putamen when assessing FC with STN.L, and decreased fluctuation in right brainstem when assessing FC with PAG.R. Compared to migraineurs, CH patients had decreased fluctuation in right brainstem and left inferior temporal lobe when assessing FC with STN.L, and decreased fluctuation in right putamen, right caudate, left middle frontal lobe and left inferior temporal lobe when assessing FC with STN.R. (Fig. 5)

Fig. 5.

Comparisons of seed-based voxel-wise FC by employing 10 selected subcortical areas. (A) Right pontine-based FC alterations in CH patients, compared to healthy controls. (B) Left spinal trigeminal nucleus-based FC alterations in CH patients, compared to controls. (C) Right periaqueductal gray-based FC alterations in CH patients, compared to controls. (D) Left spinal trigeminal nucleus-based FC alterations in CH patients, compared to migraineurs. (E) Right spinal trigeminal nucleus-based FC alterations in CH patients, compared to migraineurs. Red color represents increased connectivity while blue color represents decreased connectivity. Abbreviations: Pontine.R, right pontine; STN.L, left spinal trigeminal nucleus; STN.R, right spinal trigeminal nucleus; PAG.R, right periaqueductal gray. FWE corrected p < 0.05

In conclusion, compared to healthy controls, both CH patients and migraineurs had increased FC between SMA.R and PoCG.R, and decreased FC between PAG.L and PAG.R. In comparison with both healthy controls and migraineurs, CH patients had increased FC between PCL.L and PCL.R.

Group differences in anatomical FN values

FN analysis based on AAL116 atlas showed that, in comparison with migraineurs, CH patients had increased FN between right putamen and right cerebellum 9, and between right pallidum and right cerebellum 9 (FDR corrected p < 0.05). On the other hand, compared to migraineurs, CH patients had decreased FN between PreCG.R and right superior frontal gyrus, between PreCG.R and right inferior temporal gyrus, between left olfactory cortex and right gyrus rectus, between left cerebellum 8 and left hippocampus, between left cerebellum 8 and PHG.L, between left cerebellum 8 and right cerebellum crus1, between left cerebellum 8 and right cerebellum crus2, between left cerebellum 8 and right cerebellum 6, between left cerebellum 8 and right cerebellum 8, between left cerebellum 8 and left cerebellum 9, and between right cerebellum crus2 and right cerebellum 10 (FDR corrected p < 0.05) (Fig. 6). No significant correlations were found between FN values and patients’ clinical characteristics.

Fig. 6.

Comparisons of fiber number (FN) based on AAL116 atlas between different groups. Anatomical FN alterations in CH patients, compared to migraineurs (FDR corrected p < 0.05). Hot colors represent increased FN values while cool colors represent decreased FN values. Abbreviations: PreCG.R, right precentral gyrus; SFGdor.R, dorsolateral part of right superior frontal gyrus; OLF.L, left olfactory cortex; REC.R, right gyrus rectus; HIP.L, left hippocampus; PHG.L, left parahippocampal gyrus; PUT.R, right putamen; PAL.R, right pallidum; ITG.R, right inferior temporal gyrus; CRBLCrus1.R, right cerebellum crus1; CRBLCrus2.R, right cerebellum crus2; CRBL6.R, right cerebellum 6; CRBL8.L, left cerebellum 8; CRBL8.R, right cerebellum 8; CRBL9.L, left cerebellum 9; CRBL9.R, right cerebellum 9; CRBL10.R, right cerebellum 10

Although no significant difference in FN was found between controls and patients using FDR correction, both CH patients and migraineurs showed a tendency of increased FN compared to controls by employing NBS approach (data not shown).

The FC-FN coupling analysis were also conducted at the whole-brain level and no significant difference was observed between groups (p = 0.871 for CH vs. control, p = 0.494 for migraine vs. control, and p = 0.997 for CH vs. migraine).

Group differences in cortical thickness values

The vertex-wise analysis showed that, in comparison with healthy controls, CH patients presented reduced cortical thickness in clusters located in the left pericalcarine, right cuneus and right lateraloccipital (Monte-Carlo simulations with cluster corrected p < 0.05 and vertex-wise threshold of p = 0.001) (Fig. 7A). Compared to migraineurs, CH patients presented reduced cortical thickness in clusters located in the left pericalcarine, right pericalcarine and right lateraloccipital (Fig. 7B). No significant difference was observed between other groups or in other regions by employing cluster corrected p < 0.05 and vertex-wise threshold of p = 0.001.

Fig. 7.

Comparisons of cortical thickness values between different groups. (A) Cortical thickness alterations in CH patients, compared to healthy controls. Cool colors represent decreased cortical thickness. Monte-Carlo simulations with cluster corrected p < 0.05 and vertex-wise threshold of p = 0.001. (B) Cortical thickness alterations in CH patients, compared to migraineurs. Monte-Carlo simulations with cluster corrected p < 0.05 and vertex-wise threshold of p = 0.001. (C) Correlation between altered cortical thickness values and clinical characteristics in headache patients (corrected p < 0.05). Red colors represent positive correlation while blue colors represent negative correlation

In CH patients, Pearson partial correlation analyses showed a positive correlation between cortical thickness in left pericalcarine and disease course (r = 0.287, p = 0.013), controlling for sex and age as nuisance covariates. In headache patients (including CH and migraine), cortical thickness in left pericalcarine was positively correlated with attack duration (r = 0.291, p < 0.001), while negatively correlated with NRS score (r = -0.308, p < 0.001) (Fig. 7C). Moreover, for CH patients, a positive correlation was still observed between cortical thickness in left pericalcarine and disease course (r = 0.313, p = 0.007) while controlling for multiple clinical characteristics (sex, age, NRS score, and attack duration) as nuisance covariates. Besides, for headache patients (including CH and migraine), a positive correlation was still observed between cortical thickness in left pericalcarine and attack duration (r = 0.246, p < 0.001) while controlling for multiple clinical characteristics (sex, age, disease course, and NRS score) as nuisance covariates. In addition, for headache patients (including CH and migraine), a negative correlation was still observed between cortical thickness in left pericalcarine and NRS score (r = -0.283, p < 0.001) while controlling for multiple clinical characteristics (sex, age, disease course, and attack duration) as nuisance covariates.

Discussion

In this study, utilizing both a relatively large sample size and ultra-high-resolution images compared to prior cohorts, we identified the functional and structural alterations shared by CH and migraine, and emphasized the imaging features that differentiate CH from migraine.

Functional alterations in CH compared to controls

The first interesting findings were several significant differences in functional network patterns obtained by employing intra- and inter-network ICA. We found that CH patients in the “in bout” phase presented decreased fluctuation in AN, compared to healthy controls. Frequent nociceptive input may induce impaired supraspinal modulation in sensory-discriminatory brain areas. For CH patients in the “out of bout” phase, decreased fluctuations within the sensorimotor and the visual network were demonstrated [26]. But we detected decreased fluctuation in visual network in migraineurs compared to controls, possibly indicating some sort of similarity between CH patients in the “out of bout” phase and migraineurs. For inter-network analysis of ICs, we found that GAD-7 score was negatively correlated with the connectivity between visual network and somatomotor network. A probable hypothesis is that concerted visual and somatomotor activation may reflect patients’ capability to handle recurrent excruciating nociceptive messaging, otherwise failure to integrate them might result in increased anxiety. However, this hypothesis should be tested by more delicate studies, especially longitudinal studies enrolling patients before and after effective therapies. ICA and anatomical FC analysis are two complementary methods to estimate connectivity between large-scale brain areas [27]. Consistent with increased connectivity between sensory related ICs in CH patients compared to controls, our anatomical FC analysis based on AAL atlas found generally increased FC values between different brain regions, such as between CAL.R and PoCG.R. Increased FC values were also found between left thalamus and bilateral sensory-discriminatory areas in episodic CH patients in the “out of bout” phase [26]. For chronic CH patients, reduced FC was present in the right frontal pole-right amygdala pathway [28]. Another study showed decreased FC between bilateral thalamus and SN, which includes anterior cingulate cortex and anterior insula, in CH patients [29]. Differences in the participant enrollment (episodic vs. chronic), FC construction methods (large-scale data-driven analysis vs. selected ROI focused analysis), and more probably, sample size and MRI resolution may explain these discrepancies.

Functional alterations in migraine compared to controls

For comparison between migraineurs and healthy controls, we found that migraineurs had increased connectivity between MVN and LVN. Enhanced connectivity between visual cortex with other sensory cortices were found in migraine compared to controls [30]. Elevated entropy of the sensorimotor network was reported in migraine [31]. Our anatomical FC analysis based on AAL atlas found a tendency of elevated FC values between CAL.R and FFG.L. Aura occurs in about 1/7 to 1/3 of migraineurs, and visual symptoms are the most frequent manifestation of aura [32, 33]. The elevated FC between CAL.R and FFG.L might be an adaptive mechanism of nociceptive transmission in interictal migraineurs. Migraineurs with higher self-ratings of depression exhibited higher FC values between emotion-related areas [34]. Pontine activation was associated with the perception of pain during migraine attacks [35]. Increased activity within the dorsal pons was found during the phase of spontaneous migraine attacks, and increased FC was shown between the pons and somatosensory cortex during attacks with aura [36]. Our anatomical FC analysis based on atlas found a tendency of elevated FC values between right pontine and right superior frontal gyrus (Edge p = 0.005, uncorrected). But we did not detect aberrant FC between the pons with other brain areas. Differences in the experimental setting (interictal vs. during attacks) could explain the discrepancy between the results of these studies.

Imaging features shared by and differed between CH and migraine

ICA showed that both CH patients and migraineurs had increased connectivity between MVN and LVN. Photophobia was reported in 56% CH patients of a British cohort [37] and in 61.2% CH patients of a Germany cohort [38]. Besides, we revealed that CH patients had increased connectivity between visual network and somatomotor network, which were associated with mechanical hypersensitivity during pain perception [39]. The central processing of visual and somatomotor modalities might result in an enhanced interaction between these ICs. Compared to healthy controls, both CH patients and migraineurs presented increased FC between CAL and FFG. CAL processes visual information whereas FFG is the key region for face recognition [40]. Besides, both CH patients and migraineurs had increased connectivity between SMA.R and PoCG.R. These results indicate that brain areas mediating sensory-motor and perceptual information processing may be involved in these primary headache patients.

Compared to migraineurs, CH patients had increased FC among limbic brain regions, including PHG, cingulate gyrus and insula. The limbic system regulates pain response and controls emotions. The increased FC in CH might suggest its enhanced pain control circuit. Higher FC between cerebellar and auditory language comprehension networks were found in CH compared to migraine [41]. Our FC analysis based on atlas also detected that CH patients had a tendency of higher FC between left cerebellum 4&5 and left Heschl gyrus, and between left cerebellum 4&5 and right Heschl gyrus (Edge p = 0.01, NBS corrected component p < 0.05, iteration = 5000), compared to migraineurs. The left thalamus was shown increased FC with the cerebellum, cingulum, middle frontal gyrus and occipital areas, but decreased FC with parietal regions, middle temporal and medial frontal gyri [21]. We did not detect significant decreased FC between thalamus and cortical gyri, possibly due to the diverse manifestation of CH patients in different phases (“in bout” or “out of bout”), and our initial exclusion of cases with probable diagnosis or fulfilling concurrent diagnosis of both CH and migraine. Our structural connectivity analysis found that CH patients had increased FN between right lentiform nucleus and right cerebellum, non-overlapping with the increased FC associated regions, and generally decreased FN among cerebellar regions. Besides, no significant between-group difference was observed by employing the FC-FN coupling analysis (p > 0.05). These findings might imply that the increased FC in CH is independent of FN alterations, or at least no direct consequence of FN changes, indicating possible functional re-organization against a relatively preserved macro-structural scaffold. However, the limitation of single-shell deterministic tractography using 7T MRI data should also be considered.

The vertex-wise analysis showed that CH patients presented reduced cortical thickness in clusters located in the left pericalcarine, right cuneus and right lateraloccipital compared to controls, whereas reduced cortical thickness in clusters located in the left pericalcarine, right pericalcarine and right lateraloccipital compared to migraineurs. On the other hand, anatomical FC analysis found that CH patients had increased FC between right cuneus and PoCG.R compared to controls, and increased FC between right inferior occipital gyrus and right inferior frontal gyrus compared to migraineurs. Probably, the cortical thinning in CH patients might be related to increased FC between associated regions. However, given the cross-sectional group-level nature of the imaging findings, this speculation is only tentative. Similar to our findings, cortical thinning in collateral/lingual sulcus was found in chronic CH patients compared to controls [42]. Lower grey matter volume in inferior frontal gyrus was reported in CH patients, compared to migraineurs [41]. Furthermore, we found that cortical thickness in the left pericalcarine was positively correlated with attack duration and negatively correlated with NRS score in headache patients. Whether the cortical thinning is the cause or consequence of headache attacks should be resolved by longitudinal studies encompassing patients in different phases, or before and after medical treatment, to address whether the changes in cortical thickness are irreversible. Both CH patients and migraineurs had weaker structural covariance of hypothalamic region volume with cortical thickness [43]. The hypothalamus was presumed to be pivotal in CH lighting and hypothalamic deep brain stimulation showed effectiveness for CH management [44, 45]. Although the hypothalamus was presumed to be an important pathophysiological structure for circadian features shared by CH and migraine [46], its related connectivity might be distinct between them. But we did not detect significant between-group difference in hypothalamic FC values by employing both anatomical FC analysis and seed-based voxel-wise FC analysis, possibly suggesting other potential pathophysiological mechanisms of CH and migraine, such as generally increased connectivity between functional networks and between anatomical brain regions, which wait for further longitudinal studies to clarify.

Limitations

Our study has several limitations. First, the sex imbalance between groups, especially between CH (predominantly males) and migraine (predominantly females), is an important structural feature of this dataset. Although we analyzed every comparison between groups by controlling for sex as a nuisance covariate, residual confounding was difficult to exclude. Second, for a 7T fMRI study, rigorous control of motion and physiological noise is important. Although we applied tight but comfortable foam padding to minimize head motion, minimal framewise displacement (FD) was unavoidable. There was no significant difference in the mean FD metrics between groups (meanFD_Power = 0.1748 ± 0.0831 for CH, meanFD_Power = 0.1837 ± 0.1331 for migraine, and meanFD_Power = 0.1722 ± 0.1085 for controls). We processed fMRI data by regressing Friston 24 parameters of head motion and no scrubbing was applied. Although we removed the first volumes for participants’ adaptation and used bandpass filtering to reduce the influence of noise, the physiological signals (cardiac and respiratory) were not recorded and modelled. To further minimize possible physiological artifacts, our identified FC alterations could be validated by an independent sample. Third, we used multiple imaging features and clinical characteristics to explore their correlation. Although we take into account sex and age in the analysis, interaction effect between other metrics could not be ruled out. Further research with a much larger sample size and more stringent multiple-comparison correction is required to further clarify these findings, possibly provoking new hypothesis. Fourth, though the distribution of migraine with and without aura reflected the real-world scenario as we have reported [14], the small number of migraineurs with aura in this study rendered the comparison within migraine group limited power, thus calling for further migraine studies with enough participants to focus on this exploratory topic.

Conclusions

In this study, we revealed the MRI alterations shared by CH and migraine, and addressed the imaging feature differences between them. We also integrated the altered functional connectivity and cortical thickness in associated areas. Furthermore, we explored the correlation between imaging biomarkers and clinical characteristics. To the best of our knowledge, this study has the largest sample size with ultra-high-resolution images for CH and migraine investigation. Future studies encompassing MRI data during different periods (both “in bout” and “out of bout” phase for CH, both preictal and interictal phase for migraine) should address whether the observed image feature alterations are due to frequent excruciating attacks or a genuine pathology of CH, which could be utilized to differentiate from migraine, and probably inspiring new effective treatment for these primary headaches.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

CH

Cluster headache

ICHD-3

The third edition of international classification of headache disorders

NRS

Numerical rating scale

BOLD

Blood-oxygen-level-dependent

DTI

Diffusion tensor imaging

MNI

Montreal Neurological Institute

ICA

Independent component analysis

DMN

Default mode network

AN

Auditory network'

LFPN

Left frontoparietal network

RFPN

Right frontoparietal network

DAN

Dorsal attention network

VAN

Ventral attention network

DSMN

Dorsal somatomotor network

VSMN

Ventral somatomotor network

LVN

Lateral visual network

MVN

Medial visual network

SN

Salience network

ECN

Executive control network

FC

Functional connectivity

FN

Fiber number

AAL

Automated anatomical labelling

ROI

Region of interest

PHQ-9

Patient health questionnaire-9

GAD-7

Generalized anxiety disorder-7

PSQI

Pittsburgh sleep quality index

Author contributions

LHZ, XYW, ZD and XL contributed to the study design. LHZ, YQX, SHZ, JYH, CHD, SW, XBB, YNL, and ZXL contributed to the data acquisition. LHZ and XYW contributed to the data analysis and drafting the manuscript. All authors have read and approved the submitted version. LHZ and XYW contributed equally to this article.

Funding

This work was supported by the National Natural Science Foundation of China (No. 82327803 and 82441014 to XL, 82171208 to Zhao Dong, 81502186 to LHZ, and 82302146 to YQX).

Data availability

Data of this study are available from the corresponding author upon reasonable request.

Declarations

Ethics approval and consent to participate

The study was approved by the Ethics Committee of Chinese PLA General Hospital in accordance with the Helsinki Declaration (S2023-459-01). Written informed consent was obtained from each participant prior to the study.

Consent for publication

All authors consent for the publication.

Competing interests

The authors declare no competing interests.