Altered temporal variability-based functional reorganization of brain networks predicts motor outcome after stroke

Abstract

Background

Dynamic functional connectivity (FC) studies have shown that motor recovery after stroke was associated with functional reorganization of brain networks. However, most previous studies have focused on interregional variability rather than the temporal variability (TV) of specific regions or networks. TV quantifies the dynamic reconfiguration of a region’s or network’s functional connectivity profile over time and reflects neural flexibility.

Purpose

This study investigated functional reorganization in chronic subcortical stroke using TV of brain networks derived from resting-state fMRI.

Methods

Thirty-three patients with left subcortical stroke (LSS), thirty with right subcortical stroke (RSS), and fifty-six age- and sex-matched healthy controls (HCs) were enrolled. Stroke patients underwent resting-state fMRI and Upper Extremity Fugl-Meyer Assessment (UE-FMA) at two time points. TV was computed to characterize dynamic functional connectivity at regional, intra-network, and inter-network levels. Group differences were assessed using one-way ANCOVA with post hoc tests. Linear regression was used to examine associations between TV and motor outcomes. The false discovery rate was used to multiple comparisons correction.

Results

Compared with HCs, both LSS and RSS showed significantly reduced TV in the right frontal-cingulate regions, the somatomotor hand network (SSH), and the connections between SSH and higher-order cognitive networks (all p < 0.05, |Cohen’s d| > 0.49). Increased TV was observed in the left postcentral gyrus, inferior frontal gyrus, cerebellar network (CEN), and somatomotor mouth network (all p < 0.05, |Cohen’s d| > 0.48). Relative to LSS, RSS exhibited additional TV reductions in the right middle occipital gyrus, orbital middle frontal gyrus, default mode network (DMN), and interactions among higher-order cognitive networks (all p < 0.05, |Cohen’s d| > 0.65). Notably, TV in the right opercular inferior frontal gyrus (IFGoperc) (β = 102.69, adjusted p = 6.4 × 10^− 5) and CEN (β = 27.87, adjusted p = 0.011) at the first observation positively correlated with UE-FMA scores at follow-up, with effects modulated by lesion laterality.

Conclusion

TV captures multiscale functional reorganization in chronic subcortical stroke involving motor, cognitive, and sensory networks. TV of the right IFGoperc showed potential as a neuroimaging biomarker for predicting post-stroke motor recovery.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12984-026-02037-z.

Introduction

Stroke is a leading cause of upper-extremity motor dysfunction. Although partial recovery often occurs over time [1] and behavioral recovery tends to stabilize during the chronic phase, upper extremity motor impairment persists in 55–75% of patients with chronic stroke [2], including a high likelihood of long-term disability [3]. Improving upper extremity function is a major therapeutic target in stroke rehabilitation, as it can maximize functional recovery and reduce long-term disability [4]. Therefore, understanding the reorganization patterns of brain functional networks in chronic stroke patients may enhance the clinical outcomes in upper extremity rehabilitation. Neuroimaging studies have consistently shown that motor recovery is associated with functional reorganization of motor networks [5–8]. Other work suggests that recovery resembles a motor relearning process [9, 10], in which cognitive control systems support spontaneous recovery and rehabilitation training [11–15]. Despite this progress, the interaction between motor and non-motor systems underlying post-stroke motor recovery remains incompletely understood.

Resting-state fMRI studies demonstrated that both static and dynamic functional reorganization contributed to motor recovery after stroke [5–8]. Dynamic functional connectivity (FC) analyses have been used to identify network states and to predict outcomes in acute stroke [16, 17]. However, most studies focused on interregional variability and did not directly quantify temporal variability (TV) of specific regions or networks.

TV reflects how the FC profile of a brain region or network changes across time windows, capturing dynamic reconfiguration into different functional modules [18, 19]. Higher TV indicates greater flexibility and richer participation in multiple functional systems [18–20]. TV has been successfully applied to characterize functional reorganization in major depressive disorder [21] and Alzheimer’s disease [22]. Only one longitudinal study has examined FC variability across stroke stages, reporting reduced TV in the acute phase and increased TV in subacute and chronic stages [23]. However, the regional, intra-network, and inter-network contributions were not systematically examined, and it remains unclear whether these alterations persist in chronic stroke.

The present study aimed to characterize dynamic functional reorganization in chronic subcortical stroke with left or right hemisphere lesions using TV analysis. Based on previous reports on stroke [16, 17, 23], we hypothesized that (1) patients would show TV alterations in motor-related networks, including the sensorimotor network, fronto-parietal network, default mode network, and cerebellar network; and (2) these TV changes would correlate with upper-extremity motor function and predict recovery.

Methods

Participants

Sample size estimation was performed using G∗Power software (version 3.1.9). A pilot study was used to calculate the sample size during the study design stage. The pilot results indicated an estimated effect size (f) of 0.36 for the mean regional TV values. Assuming a two-tailed α of 0.05 and 80% power, the required sample size was 78 participants. Allowing for a 10% attrition rate, the minimum target sample size was 86 participants. To ensure adequate statistical power, more than 119 participants ( > = 30 per group) were recruited. Finally, thirty-three patients with left subcortical stroke (LSS), thirty patients with right subcortical stroke (RSS), and fifty-six HCs matched for age, sex, and handedness were recruited. To control for confounding variables and focus on mechanism of the cortical motor and non-motor networks adaption and reorganization for functional recovery [6, 24], inclusion criteria were first-ever subcortical stroke, age 27–75 years, disease duration > 3 months, and Montreal Cognitive Assessment (MoCA) score > 23. Exclusion criteria included MRI contraindications, bilateral lesions, other neurological or psychiatric disorders, unstable medical conditions, and severe aphasia. These symptoms were verbally assessed by professional clinicians (Hewei Wang and Limin Sun). Upper-extremity motor function was assessed using the UE-FMA at two time points of the first day of enrollment and 4 weeks after treatment. The study was approved by the Ethics Review Board of East China Normal University, and written informed consent was obtained from all participants.

MRI data acquisition

The MR images were scanned on a 3T MRI scanner (Siemens, Erlangen, Germany) at Shanghai Key Laboratory of Magnetic Resonance immediately after the clinical assessments. Structural images were acquired using a magnetization-prepared rapid gradient echo sequence: Repetition time (TR)/echo time (TE)/inversion time (TI) = 1900/3.42/900ms, field of view (FOV) = 240 × 240 mm², flip angle = 9°, matrix size = 256 × 256, sagittal slices = 192, slice thickness = 1 mm and slice gap = 0.5 mm. T2-weighted images were collected to identify the lesions using a turbo-spin-echo sequence: axial slices = 30, slice thickness = 5 mm, no gap, TR/TE = 6000/93 ms, FOV = 220 × 220 mm², flip angle = 120°, matrix = 320 × 320. Resting-state fMRI (rs-fMRI) data were obtained using an echo-planar imaging sequence: axial slices = 30, slice thickness = 4 mm, slice gap = 0.8 mm, TR/TE = 2000/30 ms, flip angle = 90°, FOV = 220 × 220 mm², matrix size = 64 × 64, volumes = 240 and duration time = 8 min 6 s (the first 6 s was dummy scanning). During the fMRI scans, subjects were instructed to keep motionless with their eyes closed but remaining awake yet thinking of nothing.

Lesion analysis

T1- and T2-weighted images from each stroke patient were first normalized to Montreal Neurological Institute (MNI) space. Lesions were then manually delineated slice by slice on the normalized T1-weighted images by two experienced neuroradiologists using MRIcron (www.mricro.com), with T2-weighted images used for validation. A final lesion determination was retained for each patient only if a consensus was reached between the two neuroradiologists. The resulting lesion masks were summed across all subcortical stroke (ASS) patients and within each subgroup (i.e., LSS and RSS) to generate lesion overlap maps (Fig. 1).

Fig. 1.

Lesion overlap maps for ASS, LSS, and RSS groups. For the ASS group, T1-weighted images of patients with right-hemisphere lesions were flipped to the same hemisphere. The color bar denotes the number of patients with lesions at each voxel. The z-axis ranges from 0 to 40 in MNI coordinates. ASS: all subcortical stroke; LSS: left subcortical stroke; RSS: right subcortical stroke; MNI, Montreal Neurological Institute; R, right; L, left

MRI data preprocessing

Standard analysis steps of image preprocessing were performed using two software packages: Statistical Parametric Mapping (SPM12, http://www.fil.ion.ucl.ac.uk/spm) and Data Processing Assistant for Resting-State fMRI (DPARSF, http://resting-fmri.sourceforge.net). The preprocessing procedure included the following steps: (1) removal of the first 10 volumes to allow for the steady blood-oxygenation level-dependent activity signal; (2) slice timing correction; (3) head motion correction, with exclusion of subjects who exhibited excessive motion exceeding 2.0 mm of translation or greater than 2.0 degrees of rotation in any direction. No subjects were excluded because of excessive head motion; (4) coregistration of T1 structural images to the mean functional image, followed by segmentation into different tissues (gray matter, white matter and cerebrospinal fluid); (5) spatial normalization using a diffeomorphic anatomical registration through exponentiated liealgebra (DARTEL) method, in which the lesion mask of each subject was used to remove the lesion in the MNI template space, and resampled voxel size = 3 × 3 × 3 mm³; (6) linear detrending; (7) spatial smoothing with a full width at a half maximum (FWHM) = 6 mm; (8) regression of nuisance covariates, including head motion parameters (translations, rotations, their derivatives, and quadratic terms), white matter signals, and cerebrospinal fluid signals; and (9) temporal bandpass filtering (0.01–0.1 Hz) .

Temporal variability analysis

Brain parcellation was performed using the atlas defined by Power et al., which includes 264 regions of interest (ROIs) divided into 14 functional networks [25]. To minimize potential confounding effects of subcortical lesions, 13 subcortical ROIs were excluded [24], resulting in 251 cortical ROIs for analysis. Voxel-wise time series within each ROI were averaged to generate regional signals, which were further used to derive 251 regional, 13 intra-network, and 78 inter-network time series.

TV was calculated for each brain region following the procedure described by Zhang et al. [18] (Fig. 2C). First, all blood oxygenation level dependent (BOLD) time series were divided into n non-overlapping windows, each window with a size of l, i.e. each window contained l of the 230 fMRI volumes. Within the i-th time window, the whole brain FC network F_i (a 251 × 251 matrix, for the 251 ROIs) was obtained using Pearson correlation analysis. In particular, F_i was symmetric and the diagonal elements of F_i were always 1.0. Second, the functional architecture of a region k at the i-th time window was defined as the overall FC profile of region k, which was the k-th row extracted from the matrix F_i (a 251 × 1-dimensional vector) and denoted as F_{i, k}. Third, the TV of the k-th ROI, denoted as , was defined as:

1

Fig. 2.

Schematic overview of the temporal variability calculation. A Preprocessing of BOLD time series; B Network parcellation of 251 putative functional regions based on the atlas by Power et al. [25]. C-E Computation of regional, intra-network, and inter-network temporal variability. For each time window, connectivity profiles were unfolded into vectors, and TV was calculated as one minus the average correlation coefficient of connectivity vectors across different windows. CEN, cerebellar network; DMN, default mode network; SSH, sensory/somatomotor hand; SSM, sensory/somatomotor mouth; UNC, uncertain network; AUD, auditory network; VAN, ventral attention network; VN, vision network; FPN, fronto-parietal task control network; DAN, dorsal attention network; CTC, cingulo-opercular task control network; MRN, Memory retrieval network; SAN, salience network

where, corr(,·) denoted the Pearson correlation coefficient. The was computed by comparing the FC profile of region k across different time windows i and j (i.e. all possible correlations were calculated except for i = j). This value represents the averaged the averaged correlation coefficient among different functional architecture of region k and thus a similarity measure. Next, a deduction from 1 indicates TV of a region. was initially computed across multiple window lengths (l, 20, 22, 24, …,40 volumes) and then averaged across all thevalues at these different window lengths to avoid an arbitrary choice of the window size.

The TV of FCs within a specific network is defined as intra-network variability (Fig. 2D) and the TV of FCs between two networks as inter-network variability (Fig. 2E). The computation of intra- and inter-network TV followed the same principle as that used for regional TV (Eq. 1). For a given brain network m, all FCs within this network in time window i is defined as Fm_i, which is a long vector lining up all the FCs of the network in time window i. The intra-network variability of network m then is:

2

Similarly, all FCs between networks A and B in time window i is defined as Fm_{i, A}Fm_{i, B}. The inter-network variability between networks A and B at time windows i and j is defined as:

3

Statistical analyses

Statistical analyses were performed using the Statistical Package for the Social Sciences (SPSS, version 21.0 for Windows) to compare demographic characteristics and clinical measures among the LSS, RSS, and HCs.

Baseline differences in TV among the three groups were assessed using one-way analysis of covariance (ANCOVA), followed by post hoc two-sample t-tests for pairwise group comparisons. In addition, a two-way ANCOVA was conducted on TV measures with group (RSS vs. LSS) as the between-subject factor and time (first- vs. second-observation) as the within-subject factor. When a significant group × time interaction was observed, paired or two-sample t-tests were performed for post hoc analyses. All ANCOVA models included age, sex, disease duration, lesion volume, and head motion as covariates. Multiple comparisons were corrected using the false discovery rate (FDR) method for both ANCOVA and post hoc tests, with a significance threshold of p < 0.05. It should be noted that the number of FDR corrections for each ANCOVA corresponded to the number of regional TVs (i.e., 251), intra-network TVs (i.e., 13), and inter-network TVs (i.e., 78), respectively.

Linear regression analyses were further applied to examine associations between TV measures and UE-FMA scores, controlling for age, sex, disease duration, lesion volume, and head motion. To explore the impact of lesion lateralization on the UE-FMA prediction, linear regression analyses were conducted separately for the LSS, RSS, and ASS groups at the first and second observations. FDR were used for correcting multiple comparisons with a significance threshold of p < 0.05.

To validate the robustness of the findings, all analyses were repeated using the Brainnetome Atlas [26], which comprises 210 cortical subregions and provides a fine-grained, cross-validated parcellation incorporating both anatomical and functional connectivity information.

Results

Behavioral data

Demographic characteristics and clinical assessments for stroke patients and HCs were summarized in Table 1. No significant between-group differences were observed in age, sex, lesion volume, head motion, or disease duration. In addition, UE-FMA scores differed significantly between the first and second observations in both patient groups, indicating changes in upper-extremity motor function over time.

Table 1.

Demographic and clinical data of participants

		LSS(n = 33) mean ± std	RSS(n = 30) mean ± std	HCs (n = 56) mean ± std	F/t/χ2 value	P-value
Age a (years)		55.3 ± 9.5	51.3 ± 11.0	55.3 ± 8.1	2.09	0.13
Sex b (male : female, n)		27:6	23:7	48:8	1.11	0.58
Lesion volume (ml)	Firsta	1.7 ± 2.1	2.7 ± 3.6	-	1.44	0.16
	Seconda	1.7 ± 2.2	2.6 ± 3.5	-	1.11	0.27
	First vs. Secondc t/P-value	1.24/0.22	0.83/0.42	-		-
Head motion (mm)	Firsta	0.14 ± 0.09	0.13 ± 0.09	0.11 ± 0.06	2.17	0.12
	Seconda	0.15 ± 0.12	0.14 ± 0.09		2.51	0.086
	First vs. Secondc t/P-value	0.39/0.70	0.54/0.59		-	-
Duration of illnessa (months)		8.0 ± 6.5	9.7 ± 8.1		0.93	0.35
UE-FMAa, c	Firsta	16.5 ± 13.4	17.9 ± 12.7	-	0.44	0.66
	Seconda	23.9 ± 13.4	26.3 ± 16.1	-	0.63	0.53
	First vs. Secondc t/P-value	7.64/1.1e-8***	6.97/1.2e-7***	-		-

^aANOVA for three groups and independent t-test for two groups

^bChi-Square test

^cPaired t-test. LSS: Left subcortical stroke; RSS: Right subcortical stroke; UE-FMA: Upper Extremities Fugl-Meyer Assessment; *: p < 0.05; **: p < 0.01; ***: p < 0.001

Temporal variability among HCs, LSS and RSS patients

Baseline one-way ANCOVA revealed significant between-group differences in TV across multiple brain regions, intral-networks, and inter-network connections. Regional TV differences were observed in the right opercular part of the inferior frontal gyrus (IFGoperc), left postcentral gyrus (PoCG), right supplementary motor area (SMA), bilateral paracentral lobule (PCL), right middle occipital gyrus (MOG), middle temporal pole (TPOmid), right orbital part of the middle frontal gyrus (ORBmid), anterior cingulate and paracingulate gyri (ACG), parahippocampal gyrus (PHG), orbital part of the inferior frontal gyrus (ORBinf), and left fusiform gyrus (FFG). Significant group effects were also detected at the network level, including the cerebellar network (CEN), default mode network (DMN), somatomotor hand network (SSH), and somatomotor mouth network (SSM), as well as in inter-network interactions between SSH and SSM, auditory (AUD), ventral attention (VAN), visual (VN), frontoparietal task control (FPN), and cerebellar networks. Additional inter-network differences involved DMN interactions with the uncertain network (UNC), SSM, AUD, CEN, and dorsal attention network (DAN). Post hoc analyses showed that, relative to healthy controls, the LSS group exhibited significantly increased TV in the left PoCG and IPL, accompanied by decreased TV in the right IFGoperc, SMA, PCL, ACG, and ORBinf. In contrast, the RSS group demonstrated increased TV in the left PoCG and right SMA and decreased TV in the right IFGoperc, PCL, MOG, TPOmid, ORBmid, ACG, PHG, and ORBinf, as well as in the left FFG. Furthermore, compared with the LSS group, the RSS group showed significantly higher regional TV in the right MOG, TPOmid, ORBmid, ACG, PHG, ORBinf, and left FFG (p < 0.05, |Cohen’s d| > 0.74, FDR-corrected; Table 2; Fig. 3A-C).

Table 2.

Differences in temporal variability between health controls and subcortical stroke patients

Regional temporal variability
Regional indexes of Power	Brain regions (AAL)	Hemisphere		MNI Coordinate			F value	LSS vs. HCs			RSS vs. HCs		LSS vs. RSS
				X	Y	Z		t value	Cohen’s d		t value	Cohen’s d	t value	Cohen’s d
12	IFGoperc	R		34	38	-12	14.26	-4.27***	-0.93		-4.55***	-1.02	-0.81	-0.20
27	PoCG	L		-38	-27	69	10.08	3.40**	0.74		4.47***	-1.00	-0.62	-0.16
31	SMA	R		10	-17	74	14.57	-4.17***	-0.91		-4.65***	-1.04	-0.79	-0.20
39	PCL	R		2	-28	60	9.68	-3.97***	-0.86		-3.73***	-0.84	-0.054	0.013
80	MOG	R		43	-72	28	8.19	-0.10	-0.022		-3.69***	-0.83	-3.25**	0.82
82	TPOmid	R		46	16	-30	6.85	-0.098	-0.021		-3.30**	-0.74	-3.01**	0.76
87	IPL	L		-39	-75	44	8.17	3.07**	0.67		-3.45***	0.77	-0.51	-0.13
110	ORBmid	R		8	42	-5	7.97	-0.33	-0.071		-3.78***	0.85	3.06**	0.77
122	ACG	R		12	36	20	20.09	-3.81***	-0.83		-6.16***	-1.39	-2.32*	0.58
125	PHG	R		27	-37	-13	8.79	-0.93	-0.20		-4.25**	-0.95	-2.69**	0.68
144	MOG	R		40	-72	14	6.89	-0.28	-0.061		-3.52***	-0.79	-2.58*	0.65
180	ORBinf	R		24	45	-15	22.06	-4.28***	-0.93		-5.89***	-1.33	-2.66**	0.67
235	FFG	L		-44	-65	35	10.14	-1.64	-0.36		-4.39***	-0.99	-2.44*	0.61

Intra-network temporal variability
Network
CEN							5.70	3.16**	0.69	2.50*		0.56	-0.44	-0.11
DMN							5.14	2.44*	0.53	-1.19		-0.27	-2.99**	-0.74
SSH							8.19	-2.80**	-0.61	-3.81***		-0.85	-0.91	-0.23
SSM							4.95	2.18*	0.48	2.87**		0.64	-0.76	-0.19

Inter-network temporal variability
Network 1			Network 2
UNC			DMN				6.24	0.81	0.18	-2.86**		0.64	-3.13**	-0.78
SSH			SSM				11.23	-3.34**	-0.73	-4.80***		-1.08	0.95	0.24
SSH			AUD				11.54	-3.10**	-0.67	-4.80***		-1.08	1.39	0.35
SSH			VAN				9.41	-3.11**	-0.68	-4.39***		-0.98	0.80	0.20
SSH			VN				11.76	-3.38**	-0.74	-4.49***		-1.01	-1.35	-0.34
SSH			FPN				14.34	-4.95***	-1.08	-4.16***		-0.93	-0.38	-0.094
SSH			CEN				6.31	-2.33*	-0.51	-3.48***		-0.78	0.87	0.22
SSM			DMN				5.61	0.90	0.19	-2.75**		-0.62	3.16**	0.79
AUD			DMN				5.32	-1.09	-0.24	-2.45*		-0.55	3.27**	0.81
CEN			DMN				5.19	1.43	0.31	-2.19*		-0.49	3.12**	0.78

L, left; R, right; IFGoperc, opercular part of inferior frontal gyrus; PoCG, postcentral gyrus; SMA, supplemental motor area; PCL, paracentral lobule; MOG, middle occipital gyrus; TPOmid, middle temporal pole; IPL, inferior parietal lobule; ORBmid, orbital part of middle frontal gyrus; ACG, anterior cingulate and paracingulate gyri; PHG, parahippocampal gyrus; ORBinf, orbital part of inferior frontal gyrus; FFG, fusiform gyrus; CEN, cerebellar network; DMN, default mode network; SSH, sensory/somatomotor hand; SSM, sensory/somatomotor mouth; UNC, uncertain network; AUD, auditory network; VAN, ventral attention network; VN, vision network; FPN, fronto-parietal task control network; DAN, dorsal attention network; MNI, montreal neurological institute; AAL, anatomical automatic labeling; LSS, left subcortical stroke; RSS, right subcortical stroke; HC, healthy control; *: p < 0.05; **: p < 0.01; ***: p < 0.001; FDR corrected

Fig. 3.

Significant differences in temporal variability of Power-251 regions between HCs and LSS and RSS patients, analyzed using the AAL atlas. A-C Regional temporal variability. D-F Intra-network temporal variability. G-I Inter-network temporal variability. IFGoperc, opercular part of inferior frontal gyrus; PoCG, postcentral gyrus; SMA, supplemental motor area; PCL, paracentral lobule; MOG, middle occipital gyrus; TPOmid, middle temporal pole; IPL, inferior parietal lobule; ORBmid, orbital part of middle frontal gyrus; ACG, anterior cingulate and paracingulate gyri; PHG, parahippocampal gyrus; ORBinf, orbital part of inferior frontal gyrus; FFG, fusiform gyrus; CEN, cerebellar network; DMN, default mode network; SSH, sensory/somatomotor hand; SSM, sensory/somatomotor mouth; UNC, uncertain network; AUD, auditory network; VAN, ventral attention network; VN, vision network; FPN, fronto-parietal task control network; DAN, dorsal attention network; CTC, cingulo-opercular task control network; MRN, Memory retrieval network; SAN, salience network. *: p < 0.05; **: p < 0.01; ***: p < 0.001

Regarding intra-network TV, the LSS group showed increased TV in the CEN, DMN, and SSM, and decreased TV in the SSH relative to healthy controls. The RSS group exhibited increased TV in the CEN and SSM and decreased TV in the SSH. Moreover, compared with the LSS group, the RSS group demonstrated significantly higher TV within the DMN (p < 0.05, |Cohen’s d| > 0.48, FDR-corrected; Table 2; Fig. 3D-F).

Inter-network TV analyses revealed that the LSS group showed significantly reduced TV between SSH and SSM, AUD, VAN, VN, FPN, and CEN compared with HCs. The RSS group exhibited similar reductions in these SSH-related inter-network connections, as well as additional decreases in interactions between the DMN and UNC, SSM, AUD, CEN, and DAN. Relative to the LSS group, the RSS group showed significantly lower TV in DMN-SSM, DMN-AUD, DMN-CEN, and DMN-DAN connections, but higher TV in the DMN-UNC connection (p < 0.05, |Cohen’s d| > 0.49, FDR-corrected; Table 2; Fig. 3G-I). The numbers of FDR corrections for each post-hoc test were 39 for regional TVs, 12 for intra-network TVs, and 33 for inter-network TVs.

However, the two-way ANOVA showed no significant results.

Correlation between temporal variability and UE-FMA scores

Regional TV in the right IFGoperc during the first observation of ASS (ASS_F) was positively correlated with both first UE-FMA_F (β = 85.22, adjusted p = 7.68 × 10^− 7) and second UE-FMA_S (β = 85.05, adjusted p = 5.57 × 10^− 6) scores (Fig. 4A, C). Similarly, intra-network TV in the CEN during ASS_F showed positive correlations with UE-FMA_F (β = 15.05, adjusted p = 9.06 × 10^− 4) and UE-FMA_S (β = 23.6, adjusted p = 1.83 × 10^− 4) scores (Fig. 4B, D).

Fig. 4.

Significant correlations between altered TV patterns and hemiplegic motor performance. ASS, all subcortical stroke; LSS, left subcortical stroke; RSS, right subcortical stroke; CEN, cerebellar network; IFGoperc, opercular inferior frontal gyrus; UE-FMA, Upper Extremity Fugl-Meyer Assessment Scale

In the RSS group, regional TV in the right IFGoperc was positively correlated with UE-FMAF (β = 90.77, adjusted p = 2.1 × 10^− 5) and UE-FMAS (β = 102.69, adjusted p = 6.4 × 10^− 5) scores (Fig. 4A, C). These correlations were not observed in the LSS group. Intra-network TV in the CEN in LSS showed a significant correlation with UE-FMA_S (β = 27.87, adjusted p = 0.011), but not with UE-FMA_F (β = 21.63, adjusted p > 0.05). Conversely, RSS exhibited no significant correlations between intra-network TV in the CEN and either UE-FMA_F (β = 6.08, adjusted p > 0.05) or UE-FMA_S (β = 18.12, adjusted p > 0.05) scores (Fig. 4B, D). It was worth noting that a total of 168 FDR corrections were performed in these regression analyses.

Validation results

Between-group differences in regional, intra-network, and inter-network TV among LSS, RSS, and HCs replicated the primary findings observed using the BNA atlas (Supplement file 1: Table S1). Specifically, alterations were consistently observed in regional TV of the right SMA and PCL and left ORBinf and PoCG, in intra-network TV of the sensory/somatomotor network (SMN) and default mode network (DMN), and in inter-network TV between SMN-VAN, VN-FPN, and DMN-LBN. These patterns were consistently observed across both LSS and RSS groups (Fig. 5A-C, Figs. S1-S3, Table S1). Furthermore, the correlation between IFG and UE-FMA scores remained significant, confirming the robustness of the association between altered TV and hemiplegic motor performance (Fig. 5D-E).

Fig. 5.

Validation results. A-C Significant differences in regional, intra-network, and inter-network TV of BNA regions among HCs, LSS, and RSS patients. D-E Significant correlations between altered TV patterns and hemiplegic motor performance. ASS, all subcortical stroke; LSS, left subcortical stroke; RSS, right subcortical stroke; DMN: default mode network; SMN: sensory/somatomotor network; IFGtriang, triangle inferior frontal gyrus; UE-FMA, Upper Extremity Fugl-Meyer Assessment Scale

Discussion

In the present study, we applied TV analysis to investigate functional alterations in chronic stroke patients following subcortical lesions. Our findings supported the hypothesis that subcortical stroke induced functional reorganization in both motor and non-motor regions and networks, particularly involving the IFGoperc, ACG, IPL, PoCG, CEN, DMN, and sensorimotor networks. These alterations displayed distinct patterns between RSS and LSS patients. Importantly, we observed significant positive correlations between TV in the IFGoperc and CEN at the first observation and UE-FMA scores at the second observation, suggesting that regional and intra-network TV may serve as neuroimaging biomarkers for predicting post-stroke motor recovery.

Regional temporal variability alterations in stroke groups

Compared with HCs, LSS patients showed decreased TV in the right IFGoperc, ACG, ORBinf, SMA, and PCL, and increased TV in the left PoCG and IPL. RSS patients exhibited similar changes, with additional decreases in the right MOG, TPOmid, ORBmid, and left FFG.

The increased TV in the PoCG and IPL, together with decreased TV in the SMA, align with previous reports on the roles of these regions in functional reorganization after stroke [24, 27, 28] (Supplement file 1: Discussion). Additionally, the IFG was implicated in language, cognition, emotion [29, 30], and the mirror neuron system [31], which mediated action observation [31] and motion imitation [32]. The ORBinf supported linguistic knowledge acquisition [33], whereas IFGoperc was involved in motor processes including motor imagery [34] and task execution [35]. Stimulation studies showed that enhancing SMA activity increased FC with IFGoperc, facilitating motor initiation and selection [36]. The IFGoperc is also linked to complex hand gestures [37], and subcortical lesions may disrupt connectivity between IFGoperc and SMA, contributing to motor deficits [38]. Therefore, our findings align with previous reports and further demonstrated that IFGoperc showed abnormal functional integration after stroke.

Beyond motor-related regions, we observed alterations in right MOG, TPOmid, ORBmid, and left FFG. FFG abnormalities affected visual processing [39, 40], which is critical for motor control due to the integration of visual [41], vestibular, and somatosensory inputs [42]. RSS patients showed lower FFG TV than HCs and LSS, emphasizing its role in post-stroke motor function. Structural reorganization has been reported in right MOG after chronic stroke [43], and decreased TV in right MOG in our study supports functional reorganization. Altered TV in TPOmid and ORBinf aligns with prior reports showing decreased nodal characteristics in TPOmid [44] and increased neural activities in ORBinf [8, 35]. These findings illustrate how lesion location influences brain functional reorganization and provide potential targets for rehabilitation interventions, including non-invasive neural stimulation.

Abnormal intra- and inter-network temporal variability in stroke patients

Compared with HCs, both LSS and RSS patients exhibited significantly increased TV in the SSM network and the CEN, along with decreased TV in the SSH network. The SMN plays a central role in the planning, execution, and control of voluntary movements, as well as in somatosensory and proprioceptive processing essential for perceiving body position and movement [45]. The SMN is extensively connected with other functional networks supporting higher-order cognitive processes, including attention, memory, and decision-making [46, 47]. Stroke-related damage could disrupt sensorimotor integration, impairing the transformation of sensory input into motor output and compromising goal-directed movements. Dynamic FC studies have identified three distinct SMN states in stroke patients: a state with strong intra-network but weak inter-network connectivity, a state with globally weak intra-network connectivity, and a hybrid state combining features of both [16, 17]. Further, the SMN comprises SSH and SSM subnetworks [25]. SSH primarily supported hand motor planning and execution [48, 49], whereas SSM was associated with oral sensorimotor control involved in eating, drinking, and speaking, as well as vivid oral sensations [50], the SSM exhibited high functional similarity, consistent with the strong segregation of the sensorimotor system in the human cortex [25]. Therefore, altered intra-network TV in SSH observed in the present study was consistent with previous findings [27, 51], suggesting an unstable FC architecture and increased dynamic interactions between subnetworks. Moreover, prior evidence indicated that cortical regions constituting the SSM readily reorganized in response to minimal stimulation [52–54]. Therefore, these results underscored the important role of SSM in functional reorganization driven by cortical injury following subcortical stroke.

Disruptions in neuronal activity and network connectivity within cerebellar regions were well documented after stroke and were closely associated with disease severity [55]. Dynamic FC studies have reported higher switching rates in the CEN in stroke patients compared with HCs [30]. Our findings extended this evidence by demonstrating that the cerebellum served as a critical hub for functional reorganization and that CEN alterations occurred following subcortical stroke irrespective of lesion hemisphere. Notably, functional reorganization within the CEN, SSH, and SSM exhibited a globally consistent pattern across LSS and RSS patients.

At the inter-network level, both LSS and RSS patients showed significantly decreased TV between SSH and SSM, AUD, VAN, and FPN networks compared with HCs. Higher inter-network TV was generally associated with richer FC patterns and greater cognitive flexibility [18]. The FPN was considered a highly dynamic and adaptable network capable of supporting diverse task demands [56]. Reduced TV between SSH and FPN was consistent with prior fMRI studies showing joint recruitment of these networks during motor tasks in stroke patients [47, 57]. Additionally, decreased inter-network TV between SSH and AUD supported the involvement of auditory-motor coupling in chronic stroke, as previously reported [58, 59]. This coupling relied on rapid feedforward and feedback loops that precisely coordinate auditory and motor information [60], suggesting that AUD-FPN-SSH interactions may serve as potential biomarkers in non-acute stroke. Attention networks such as VAN were cross-modal and supported dynamic integration of executive processes [61]. Given that effective motor rehabilitation depended on intact visuospatial, language, and attentional functions, our findings were consistent with evidence that multiple functional systems synchronize during post-stroke recovery.

We further observed that LSS patients exhibited significantly increased intra-network TV within the DMN compared with HCs, while RSS patients showed even greater increases relative to LSS. Despite growing interest, the functional integrity of the DMN after stroke remains incompletely understood. Existing studies suggest that disruptions in intra- and inter-network DMN connectivity are associated with motor impairments [27, 62], and increased DMN switching rates have been reported using dynamic FC approaches [63]. However, investigations into how lesion location influenced DMN reorganization were limited. Our findings provided new evidence supporting lesion lateralization effects [64] and suggest that intra-network TV was a powerful metric for characterizing DMN reorganization in chronic stroke.

Finally, RSS patients exhibited significant alterations in inter-network TV between the DMN and UNC, SSM, AUD, CEN, and DAN compared with both HCs and LSS patients. Impaired switching between the DMN and motor-related networks has been documented in chronic stroke, reflected by insufficient DMN deactivation during activation of ipsilesional motor networks [65]. Consistent with our previous work [27], these results supported the notion that higher-order cognitive networks, including the DMN and attention systems, exerted top-down modulation over motor networks to facilitate motor recovery. Moreover, our findings demonstrated that lesion lateralization influences inter-network reorganization [64], indicating that inter-network TV may serve as a valuable biomarker for characterizing functional reorganization in chronic stroke patients.

Association of functional reorganization with UE-FMA

Previous studies have demonstrated that the cerebellum played a critical role in post-stroke recovery. Neuroimaging studies have shown that a lateralized, near-normal pattern of cerebellar activation was associated with better motor performance [66–68], while FC analyses have reported increased cortico-cerebellar coupling during spontaneous recovery and neuro-rehabilitative training [7, 8]. Extending these longitudinal findings, the present study indicated that baseline functional metrics quantified early after stroke may capture meaningful inter-individual variability related to subsequent motor outcomes. In addition, a negative association has been reported between lesion size in the IFGoperc and hand action performance in chronic stroke patients [69], further highlighting the relevance of IFGoperc involvement in motor dysfunction. Using a longitudinal dataset encompassing patients with different subcortical lesion locations, we found that TV in the right IFGoperc and CEN at the first assessment in ASS_F was positively correlated with both UE-FMA_F and UE-FMA_S motor scores. These results suggest that TV has predictive value for motor recovery after stroke. Notably, lesion lateralization modulated these associations. In the LSS group, TV in the right IFGoperc showed no significant correlation with either UE-FMA_F or UE-FMA_S, whereas in the RSS group, significant positive correlations were observed with both measures. Similarly, TV in the CEN was positively correlated with UE-FMA_S, but not UE-FMA_F, in LSS patients, while no significant correlations were detected between CEN TV and UE-FMA scores in RSS patients. Collectively, these findings indicated that the predictive value of TV was influenced by lesion lateralization. Nevertheless, they supported the notion that TV in the right IFGoperc and CEN represented a promising neuroimaging biomarker for predicting motor outcomes following subcortical stroke.

Limitation

There are some limitations in the present study. First, lesion-specific mechanisms of reorganization differ between cortical and subcortical stroke. Standard motor circuitry is typically recruited after subcortical stroke, whereas alternative networks are more commonly recruited after cortical stroke [70–72]. For example, the contralateral motor cortex, ipsilateral cerebellum relative to the moving limb, bilateral mesial regions, including the cingulate cortex and supplementary motor area, and perisylvian regions are activated in subcortical stroke. In contrast, patients with cortical stroke mainly recruit ipsilateral postcentral mesial hemispheric regions and areas surrounding the stroke cavity [70]. Therefore, we restricted the present study to patients with a single lesion type, namely subcortical lesions involving similar basal ganglia regions. As a result, the conclusions of this study may not be generalizable to patients with cortical stroke. Future studies should include patients with cortical stroke to obtain more generalizable conclusions. Second, the total sample size of stroke patients is relatively small, we will expand the sample size of stroke patients in the future. Finally, although the present study did not record additional clinical data, such as cognitive assessments or information on residual neglect, aphasia, or other neurological deficits, patients with obvious symptoms were excluded by clinicians through verbal consultation. Future studies should record these additional clinical data to provide a more comprehensive clinical characterization of the participants.

Conclusion

This study demonstrates that TV in functional brain networks effectively captures dynamic reorganization in non-acute subcortical stroke patients. Significant alterations were observed in motor regions, including the right IFGoperc, SMA, and PoCG, as well as cognitive areas such as the ACG and IPL. TV changes showed positive correlations with motor recovery measured by UE-FMA scores, and lesion laterality modulated these patterns, highlighting hemispheric influences on functional adaptation. These findings establish TV as a robust predictive biomarker for motor outcomes, offering valuable insights into post-stroke neural plasticity and informing targeted rehabilitation strategies. Overall, this work advances our understanding of brain network flexibility and its clinical relevance in stroke recovery.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Conceptualization, H.W., Z.Z. and L.S.; data curation, Q.L., H.W. and M.F.; formal analysis, L.L. and M.F.; funding acquisition, H.W. and L.S.; investigation, G.X.; methodology, G.X., L.L. and Q.L.; project administration, L.S., Z.Z. and H.W.; software, M.F.,H.W. and G.X.; supervision, H.W., Z.Z.and L.S.; validation, L.L. and Q.L.; visualization, G.X.; writing–original draft, G.X.; writing–review and editing, H.W., Z.Z., L.S. and G.X. All authors have read and agreed to the published version of the manuscript.

Funding

This project was supported by grants from Research Personnel Cultivation Programme of Zhongda Hospital Southeast University (CZXM-GSP-RC182), the National Natural Science Foundation of China (81974356 and 82350002), Key Project of Jiangsu Province’s Key Research and Development Program (No.BE2023023-4), Key Project of Jiangsu Province’s Key Research and Development Program (BE2023034), Key Supporting Discipline Construction Project of Shanghai Health System (2023ZDFC0304), Shanghai Natural Science Foundation (23ZR1408500), Joint Foundation for Scientific and Technological Innovation of Fujian Province (2025Y9138) and Natural Science Foundation of Fujian Province (2025J01724).

Data availability

The data that support the findings of this study are available from Huashan Hospital but restrictions apply to the availability of these data, which were used under license for the current study, and so are not publicly available. Data are however available from the authors upon reasonable request and with permission of corresponding author.

Declarations

Ethics approval and consent to participate

The Huashan Hospital Institutional Review Board (Registration number: ChiCTR-TRC-08003005) approved the study protocol in accordance with the principles of the Declaration of Helsinki. All participants understood the study procedure and provided written informed consent before participation.

Informed consent

Written informed consent was obtained from all study participants for data collection, analysis, and publication.

Competing interests

The authors declare no competing interests.