Lesion Network Mapping for Atrial Fibrillation Newly Detected on ECG or Cardiac Monitoring After Stroke

Abstract

Background

Atrial fibrillation detected after stroke is a distinct clinical entity that may stem from cardiogenic‐neurogenic interactions. We aimed to illuminate the lesion network mapping of atrial fibrillation newly detected on ECG or cardiac monitoring after stroke and explore the association between the central autonomic network and occurrence of atrial fibrillation.

Methods

We performed voxel‐based lesion‐symptom mapping, structural disconnection connectome mapping and functional disconnection connectome mapping to locate lesions and networks for atrial fibrillation newly detected on ECG or cardiac monitoring after stroke. We also calculated the mapping score to quantify the overlap between lesions and maps and evaluated its association with atrial fibrillation using logistic regression analysis.

Results

Among 4629 patients, 149 (3.22%) had atrial fibrillation newly detected on ECG or cardiac monitoring after stroke. SDC maps revealed notable disconnections in right insular cortex, amygdala and so on with the peak Z score in the right precentral gyrus (Z score=3.27). Functional disconnection connectome maps revealed pronounced functional disconnections in bilateral precentral/postcentral gyrus, insular cortex and so on with the peak Z score in the right insular cortex (Z score=3.41). Region of interest analysis showed that functional/structural disconnections in some sympathetic/parasympathetic regions (eg, primary motor cortex, temporal pole) were associated with the occurrence of atrial fibrillation.

Conclusions

Significant structural disconnection connectome/functional disconnection connectome mapping and central autonomic network associations were observed in patients with atrial fibrillation newly detected on ECG or cardiac monitoring after stroke, supporting the crucial role of brain networks especially the central autonomic network in the pathogenesis of atrial fibrillation detected after stroke.

Nonstandard Abbreviations and Acronyms

CAN

central autonomic network

CLSM

connectome‐based lesion‐symptom mapping

FDC

functional disconnection connectome

FDR

false discovery rate

NIHSS

National Institutes of Health Stroke Scale

ROI

region of interest

SDC

structural disconnection connectome

VLSM

voxel‐based lesion‐symptom mapping

Clinical Perspective.

What Is New?

• Significant structural and functional disconnections in brain networks were identified in patients with atrial fibrillation newly detected on ECG or cardiac monitoring after stroke.

• Associations between impairment of the central autonomic network and atrial fibrillation newly detected on ECG or cardiac monitoring after stroke were verified.

What Are the Clinical Implications?

• Findings provide neuroanatomical evidence pertaining to the pathogenesis of atrial fibrillation newly detected on ECG or cardiac monitoring after stroke, suggesting potential targets for neuromodulation therapies.

Approximately 10% to 20% of patients with acute ischemic stroke experience severe adverse cardiac events, and patients with stroke with early severe cardiac complications have a 2 to 3 times increased risk of short‐term mortality. ¹ Poststroke cardiovascular events are called stroke‐heart syndrome; severe cardiac arrhythmia is an important type among them. ² The incidence of atrial fibrillation (AF) is 8‐fold higher in patients with acute ischemic stroke compared with matched individuals without a predisposition to stroke. AF detected after stroke has been proposed as a concept to characterize previously unknown but newly detected AF after stroke. ³

The central autonomic network (CAN) regulates the flow of sympathetic and parasympathetic nerves to the heart and acute ischemic stroke may lead to sudden changes in the physiological tissues of CAN leading to autonomic dysfunction. Areas of CAN that have an impact on cardiac function include the amygdala, anterior cingulate cortex, ventromedial prefrontal cortex, insula, hypothalamus, mediodorsal thalamus, hippocampus, and brainstem regions. ⁴ Previous studies have found that the lesions of right insula, right parietal cortex, right frontal lobe, right amygdala, basal ganglia, and thalamus were significantly correlated with the occurrence of arrhythmia after acute ischemic stroke. ⁵ Besides the ischemic lesions, remote brain networks that are connected to the lesions can also result in physiological changes; however, the relationship between the brain network especially the central autonomic network and the occurrence of AF newly detected on ECG or cardiac monitoring after stroke is still not clear. To address this issue, we aimed to use voxel‐based lesion‐symptom mapping, structural disconnection mapping, functional disconnection mapping, and regions of sympathetic and parasympathetic nervous system network analysis to identify the disconnection network and evaluate whether lesions in brain regions associated with CAN demonstrate significant association with the occurrence of AF newly detected on ECG or cardiac monitoring after stroke.

METHODS

Study Design and Participants

This study conducted exploratory analyses using data from the CNSR‐III (Third China National Stroke Registry), a nationwide, multicenter, prospective, observational registry study of patients with acute ischemic stroke or transient ischemic attack enrolled at 201 hospitals in China between August 2015 and March 2018. ⁶ The data that support the findings of this study are available from the corresponding author upon reasonable request. Patients underwent the 12‐lead ECG and further≥24‐hour Holter cardiac rhythm recording/telemetry at admission or during the hospitalization based on the discretion of the treating physician. Based on their history of AF and the results of ECG monitoring during hospitalization, no history of AF but detected AF during hospital stay via 12‐lead ECG or prolonged cardiac monitoring after the index ischemic stroke were defined as AF newly detected on ECG or cardiac monitoring after stroke. This study was approved by the Institutional Review Board of Beijing Tiantan Hospital and all participating centers, and written informed consent was obtained from all participants or their representatives.

The exclusion criteria included patients with following criteria: diagnosed with transient ischemic attack or without infarction on imaging; received reperfusion therapy (including thrombolysis or thrombectomy); history of stroke, cancer, dementia, epilepsy, and mental illness; a complication of hemorrhagic transformation; lacking imaging data or diffusion‐weighted imaging sequences; or unqualified image quality. Considering the effects of the other cardiogenic factors for AF newly detected on ECG or cardiac monitoring after stroke and to untangle the impact of the CAN from cardiogenic factors, we excluded patients with high‐risk sources of cardioembolism based on the TOAST (Trial of Org 10172 in Acute Stroke Treatment) criteria, ⁷ including mechanical prosthetic valve, mitral stenosis, left atrial/atrial appendage thrombus, sick sinus syndrome, recent myocardial infarction (<4 weeks), left ventricular thrombus, dilated cardiomyopathy, akinetic left ventricular segment, atrial myxoma, and infective endocarditis. A total of 4629 patients were included (Figure S1). Voxel‐based lesion symptom mapping (VLSM), structural disconnection connectome (SDC) mapping, functional disconnection connectome (FDC) mapping, and region of interest (ROI) analysis were used to identify distinct lesion and network localizations with the occurrence of AF newly detected on ECG or cardiac monitoring after stroke.

Imaging Techniques

In the CNSR‐III cohort, brain images including T1 weighted, T2 weighted, fluid‐attenuation inversion recovery, diffusion‐weighted imaging, and apparent diffusion coefficient scanning, magnetic resonance angiography, susceptibility‐weighted imaging or computed tomography (if contraindicated to magnetic resonance imaging [MRI]) in Digital Imaging and Communications in Medicine format were collected from 201 hospitals and analyzed by the image research center of Beijing Tiantan Hospital.

Voxel‐Based Lesion‐Symptom Mapping

The lesion images were registered to the Montreal Neurological Institute template‐152 using the FMRIB Software Library (version 6.0.0, https://fsl.fmrib.ox.ac.uk/fsl). An experienced investigator who was blinded to clinical parameters and physiologic parameters delineated the boundaries of the ischemic lesion on anonymized imaging scans using MRIcron (http://www.mccauslandcenter.sc.edu/mricro/mricron/). Initially, the lesion images from the diffusion‐weighted imaging space were transformed to the T1 template. The T1 images were aligned to the Montreal Neurological Institute‐152 template with a 2 mm resolution using affine alignment, generating affine transformation matrices from the individual space to the template. For lesion images in the template space, voxels marked as lesions in fewer than 5% of patients were excluded to avoid bias from sporadic voxels.

Structural and Functional Disconnection Mapping

We applied the connectome‐based lesion‐symptom mapping (CLSM) to assess the relationships between AF newly detected on ECG or cardiac monitoring after stroke and connectome pathways, and SDC and FDC maps reflecting putative network disconnections caused by infarct lesions. CLSM was also referred to as lesion network mapping and had been used in several prior studies, including network localization of addiction, cognitive performance, and the National Institutes of Health Stroke Scale (NIHSS). ⁸ , ⁹ , ¹⁰ This method used a common connectome data set to analyze the network connectivity distribution of lesion locations in patients, thereby indirectly assessing the network regions affected by the lesion. In this study, the common connectome dat aset was from an ongoing prospective community cohort study. ¹¹ This database collected resting‐state functional MRI and diffusion MRI, enabling the construction of functional and structural connectome data sets, respectively (Figure 1). This study used a subset of the cohort. The screening criteria for this subset included the exclusion of participants lacking imaging information, those with failed imaging preprocessing, and individuals with histories of stroke, heart disease, diabetes, hypertension, hyperlipidemia, or Montreal Cognitive Assessment scores <26. After screening, 142 subjects remained out of 3067 subjects screened, with a mean age of 57.42 years and a female proportion of 53.9%.

Figure 1. Lesion location and network mapping.

VLSM analysis was not able to identify associations between damage to a specific brain region and atrial fibrillation newly detected on ECG or cardiac monitoring after stroke. CLSM was then implemented to evaluate the functional and structural connectivity between each lesion location and all other brain voxels, using the connectome data from an ongoing prospective community cohort study. AF indicates atrial fibrillation; CLSM, connectome‐based lesion‐symptom mapping; FDC, functional disconnection connectome; SDC, structural disconnection connectome; and VLSM, voxel‐based lesion‐symptom mapping.

The SDC maps were generated using diffusion tensor imaging. In the TrackVis software (http://trackvis.org/), each lesion image in the template space was used as a seed for fiber tracking, generating 142 structural disconnection indicator maps. These 142 images were summed, then divided by the total number to produce a single structural disconnection map that reflects the probability of structural disconnection.

The FDC maps were generated using functional MRI. We used 198 functional images with 3 mm resolution from the same healthy control subjects to calculate the Pearson correlation between the mean signal of the lesion images in the template space and the whole brain, thereby obtaining 198 functional disconnection maps. We then conducted a 1‐sample t test for each voxel based on the 198 functional disconnection maps, using the t‐value as the final FDC map.

ROI Analysis

Based on the computed SDC and FDC maps, we conducted a ROI study. We applied the coordinates of the sympathetic nervous system network, parasympathetic nervous system network, and default mode network provided by previous studies, selecting regions within a 5 mm radius as the ROI. ⁴ , ¹² We also used the automated anatomical labeling atlas for ROI analysis. For each patient and each ROI, we extracted the mean values from their SDC and FDC maps. Subsequently, we constructed a linear model with AF newly detected on ECG or cardiac monitoring after stroke as the dependent variable and age, sex, and lesion volume as covariates, to analyze the association between the degree of structural or functional disconnection in the ROI and AF newly detected on ECG or cardiac monitoring after stroke.

Calculation of the Mapping Score

After obtaining the CLSM maps, we calculated the mapping scores of each patient's lesions with the 3 maps. The calculation method for these scores is consistent with a previous study. ⁹ In simple terms, the mapping score for each patient was obtained by summing the values of the CLSM maps within their lesion areas. Furthermore, based on previous research, ¹³ we calculated the quartiles for all patients' scores and reassigned the map scores accordingly. Ultimately, for each map result, each patient received a score ranging from 1 to 4. Higher values indicate a greater extent of the patient's lesion falling within the mapping atlas.

Statistical Analysis

All statistical analyses were performed using MATLAB 2019b (MathWorks Inc., Natick, MA). Participants were divided into 2 groups: those with AF newly detected on ECG or cardiac monitoring after stroke and those without based on medical history and ECG/Holter monitoring results during hospitalization. For univariate analysis of baseline clinical data, normality of data distribution was assessed using the Shapiro–Wilk test. Continuous variables with normal distribution were expressed as mean±SD and compared using independent samples t tests. Nonnormally distributed variables were presented as median (interquartile range) and analyzed by Wilcoxon rank‐sum tests. Categorical variables were described as counts (percentages) and compared using χ² tests or Fisher's exact tests, as appropriate. Statistical significance was set a priori at P≤0.05.

To compare lesion frequency at each voxel between patients those with AF newly detected on ECG or cardiac monitoring after stroke and those without, and to identify brain regions significantly associated with AF newly detected on ECG or cardiac monitoring after stroke, this study performed VLSM analysis. A logistic regression model was conducted for each voxel to analyze the relationship between lesion presence and AF newly detected on ECG or cardiac monitoring after stroke, with age, sex, NIHSS score, lesion volume and cardiogenic factors (including anterior–posterior diameter of left atrium, left ventricular ejection fraction, and regional wall movement abnormalities) as covariates. Multiple comparisons were corrected using the false discovery rate (FDR), and regions with an FDR‐corrected P value of <0.05 were considered significant.

We conducted CLSM analyses to assess the relationships between AF newly detected on ECG or cardiac monitoring after stroke and connectome pathways. The SDC and FDC maps reflect a putative network disconnection caused by the infarct lesion of AF newly detected on ECG or cardiac monitoring after stroke. A linear regression model was conducted for each voxel to analyze the relationship between network disconnection and AF newly detected on ECG or cardiac monitoring after stroke, with age, sex, NIHSS score, lesion volume, and cardiogenic factors as covariates. For the CLSM analysis results, multiple comparisons were similarly corrected using the FDR with corrected P value <0.05. To further investigate the association between involvement of CAN‐related brain regions and the occurrence of AF newly detected on ECG or cardiac monitoring after stroke, we conducted ROI analyses using a series of linear regression models, specifically targeting key neural network nodes within the sympathetic and parasympathetic systems. We reported the positive results of the statistical model, indicating that the degree of structural or functional network disconnection in patients with AF newly detected on ECG or cardiac monitoring after stroke is greater than that in patients without AF newly detected on ECG or cardiac monitoring after stroke.

To analyze the relationship between mapping scores and AF newly detected on ECG or cardiac monitoring after stroke, we first applied nonparametric tests to compare score differences between patients with AF newly detected on ECG or cardiac monitoring after stroke and those without. Subsequently, we performed logistic regression models with the mapping scores of the three maps and AF newly detected on ECG or cardiac monitoring after stroke, using age, sex, NIHSS score, lesion volume, and cardiogenic factors as covariates.

RESULTS

Patient Characteristics

Among 4629 patients, 149 (3.22%) had AF newly detected on ECG or cardiac monitoring after stroke. When comparing characteristics between the 2 groups, patients with AF newly detected on ECG or cardiac monitoring after stroke were older (71 versus 63, P<0.001) and they had higher NIHSS scores on admission (5 versus 3, P<0.001) and larger lesion volume (18.17 mL versus 4.69 mL, P<0.001). Besides, patients with AF newly detected on ECG or cardiac monitoring after stroke had higher prevalence of coronary artery disease (18.79% versus 9.46%, P<0.001), and myocardial infraction (4.03% versus 1.61%, P=0.02) (Table 1). Among the 4629 analyzed patients, stroke subtype distribution according to the TOAST criteria was as follows: large‐artery atherosclerosis in 1343 cases (29.01%), cardioembolism in 156 (3.37%), small‐artery occlusion in 1218 (26.31%), stroke of other determined cause in 69 (1.49%), and stroke of undetermined cause in 1843 (39.81%). Comparative analysis of baseline characteristics between patients included in the final analysis and those excluded due to missing ECG/Holter data was shown in Table S1. Patients excluded had slightly higher NIHSS scores (P=0.02) and lower body‐mass index (P<0.001). All other demographic and clinical variables, including age, sex, lesion volume, and medical history, showed no significant differences (Table S1).

Table 1.

Demographic and Clinical Characteristics of the Patients

Characteristics	Total population (n=4629)	No‐atrial fibrillation newly detected on ECG or cardiac monitoring after stroke (n=4480)	Atrial fibrillation newly detected on ECG or cardiac monitoring after stroke (n=149)	P value
Age, y, median (IQR)	63 (52–72)	63 (52–72)	71 (63–78)	<0.001
Female, n (%)	1430 (30.89)	1377 (30.74)	53 (35.57)	0.21
National Institutes of Health Stroke Scale score at admission, median (IQR)	3 (1–6)	3 (1–6)	5 (2–10)	<0.001
Body‐mass index, kg/m2, mean ±SD	24.86±3.34	24.87±3.33	24.60±3.47	0.33
Lesion volume, mL, median (IQR)	4.81 (1.51–21.04)	4.69 (1.49–19.45)	18.17 (3.08–89.87)	<0.001
Medical history, n (%)
Hypertension	2978 (64.33)	2880 (64.29)	98 (65.77)	0.71
Diabetes	1077 (23.28)	1045 (23.33)	32 (21.48)	0.60
Dyslipidemia	370 (7.99)	360 (8.04)	10 (6.71)	0.56
Current tobacco smoker	1499 (32.38)	1466 (32.72)	33 (22.15)	0.007
Current drinker	776 (16.76)	758 (16.92)	18 (12.08)	0.12
Coronary artery disease	452 (9.76)	424 (9.46)	28 (18.79)	<0.001
Heart failure	22 (0.48)	20 (0.45)	2 (1.34)	0.12
Myocardial infarction	78 (1.69)	72 (1.61)	6 (4.03)	0.02
Peripheral vascular disease	33 (0.71)	60 (1.34)	1 (0.69)	0.35
Stroke	910 (19.66)	722 (16.11)	29 (19.78)	0.27
Transient ischemic attack	113 (2.44)	110 (2.46)	3 (2.01)	0.73

Data are mean±SD, n (%), or median (IQR). IQR indicates interquartile range.

Imaging Characteristics

All patients enrolled in this study underwent 1.5T or 3T MRI scans within 7 days of onset. The distribution of infarct lesions based on voxels (Figure 2) suggests that lesions in the enrolled patients were involved in both cortical and subcortical areas, and the left and right hemispheres were almost symmetrical.

Figure 2. Lesion overlap and distribution map of atrial fibrillation newly detected on ECG or cardiac monitoring of patients after stroke following acute ischemic stroke.

Regions where lesions are present in at least 2% of 4629 patients are shown. Lesion frequency increases from dark red to yellow. L: Left hemisphere, N: Percentage of samples with lesions at the corresponding voxel location, R: Right hemisphere.

Lesions and Network Localizations of Atrial Fibrillation Newly Detected on ECG or Cardiac Monitoring After Stroke

After controlling for age, sex, NIHSS score, lesion volume, and cardiogenic factors as covariate, VLSM analysis did not identify any significant differences after FDR correction. Voxel‐based analysis of the SDC revealed that structural disconnection in the right insular cortex, right amygdala, right parahippocampal gyrus, right inferior fronto‐occipital fasciculus, and right inferior longitudinal fasciculus is significantly associated with the occurrence of AF newly detected on ECG or cardiac monitoring after stroke (Figure 3A). The peak Z score predicting AF newly detected on ECG or cardiac monitoring after stroke of the SDC was found in the right precentral gyrus (peak coordinates x=52, y=−2, z=18, Z score=3.27). Analysis of the FDC maps indicated that AF newly detected on ECG or cardiac monitoring after stroke is significantly associated with functional disconnection in the bilateral precentral gyrus, bilateral postcentral gyrus, bilateral insular cortex, bilateral operculum cortex, and bilateral amygdala (Figure 3B). The peak Z score predicting AF newly detected on ECG or cardiac monitoring after stroke of the FDC was found in the right insular cortex (peak coordinates x=39, y=−9, z=6, Z score=3.41). The detailed information on the distribution of SDC and FDC maps within various brain regions was reported in Table S2.

Figure 3. CLSM analysis of atrial fibrillation newly detected on ECG or cardiac monitoring of patients after stroke.

A, Structural disconnection regions in patients with atrial fibrillation newly detected on ECG or cardiac monitoring after stroke. B, Functional disconnection regions in patients with atrial fibrillation newly detected on ECG or cardiac monitoring after stroke. All results were FDR corrected by the threshold of 0.05. The statistical t‐value increases from dark red to yellow, indicating that the disconnection in the patients with atrial fibrillation newly detected on ECG or cardiac monitoring after stroke is significantly greater than that in the patients without atrial fibrillation newly detected on ECG or cardiac monitoring after stroke. CLSM indicates connectome‐based lesion‐symptom mapping; and FDR, false discovery rate.

Additionally, we performed a sensitivity analysis by including the 35 patients who were initially excluded due to high risk of cardioembolic cause. VLSM analysis also did not identify significant differences after FDR correction. Voxel‐based analysis of the SDC revealed that structural disconnection in the right insular cortex, right amygdala, right parahippocampal gyrus, right inferior fronto‐occipital fasciculus, and right uncinate fasciculus is significantly associated with the occurrence of AF newly detected on ECG or cardiac monitoring after stroke (Figure S2A), while the peak z‐score was found in the right frontal orbital cortex (peak coordinates x=24, y=16, z=−22, Z score=3.34). Analysis of the FDC maps indicated that AF newly detected on ECG or cardiac monitoring after stroke is significantly associated with functional disconnection in the bilateral insular cortex, bilateral temporal cortex, bilateral operculum cortex, and bilateral amygdala (Figure S2B), the peak Z score of the FDC was found in the left insular cortex (peak coordinates x=−39, y=−9, z=3, Z score=5.44). The results demonstrate similar findings with our primary analysis.

ROI Analysis

ROI analysis of structural disconnection revealed that no node is associated with AF newly detected on ECG or cardiac monitoring after stroke. Functional disconnection in 2 parasympathetic network nodes: bilateral primary motor cortex, temporal pole (left t‐value=2.85, right t‐value=2.90); 2 sympathetic network nodes: bilateral supramarginal gyrus, superior parietal lobule, primary somatosensory cortex (left t‐value=2.71, right t‐value=2.64); and 12 automated anatomical labeling brain regions: right precentral gyrus (t=2.61), right inferior frontal gyrus–opercular part (t=2.61), left rolandic operculum (t=2.76), right rolandic operculum (t=3.62), left postcentral gyrus (t=2.98), right postcentral gyrus (t=2.78), left supramarginal gyrus (t=3.33), right supramarginal gyrus (t=3.02), left Heschl's gyrus (t=2.68), right Heschl's gyrus (t=3.26), left superior temporal gyrus (t=2.76), and right superior temporal gyrus (t=3.05) are associated with AF newly detected on ECG or cardiac monitoring after stroke (Figure 4). The involved nodes, t values, P values, and coordinates in Montreal Neurological Institute space were demonstrated in Table 2.

Figure 4. The result of regions of interest analysis in patients with atrial fibrillation newly detected on ECG or cardiac monitoring after stroke.

Node colors represented their types, and the sizes indicated the strength of difference. AAL indicates automated anatomical labeling.

Table 2.

Brain Areas Associated With Sympathetic and Parasympathetic Regulation as Assessed by Atrial Fibrillation Newly Detected on ECG or Cardiac Monitoring After Stroke

	Involved nodes	ROI	L/R	t‐value	FDR corrected P value	x	y	z
Functional disconnection connectome	Primary motor cortex, temporal pole	Parasympathetic	L	2.85	0.03	−56	6	8
	Primary motor cortex, temporal pole	Parasympathetic	R	2.90	0.03	56	6	8
	Supramarginal gyrus, superior parietal lobule, primary somatosensory cortex	Sympathetic	L	2.71	0.04	−44	−36	42
	Supramarginal gyrus, superior parietal lobule, primary somatosensory cortex	Sympathetic	R	2.64	0.04	44	−36	42
	Precentral	AAL	R	2.61	0.04	50	6	22
	Frontal_inf_oper_R	AAL	R	2.61	0.04	49	15	21
	Rolandic_oper_L	AAL	L	2.76	0.03	−48	−8	14
	Rolandic_oper_R	AAL	R	3.62	0.004	48	−8	14
	Postcentral_L	AAL	L	2.98	0.02	−43	−24	47
	Postcentral_R	AAL	R	2.78	0.03	43	−24	47
	SupraMarginal_L	AAL	L	3.33	0.009	−57	−34	30
	SupraMarginal_R	AAL	R	3.02	0.02	57	−34	30
	Heschl_L	AAL	L	2.68	0.04	−45	−17	10
	Heschl_R	AAL	R	3.26	0.01	45	−17	10
	Temporal_sup_L	AAL	L	2.76	0.03	−54	−21	7
	Temporal_sup_R	AAL	R	3.05	0.02	54	−21	7

AAL indicates anatomical automatic labeling; FDR, false discovery rate; L, left; R, right; and ROI, region of interest.

Mapping Score Analysis

As shown in Figure 5A, we calculated the degree of overlap between each lesion image and the FDC and SDC maps. All samples with an overlay value >0 were assigned scores ranging from 1 to 4 based on quartile classification. A higher score indicated that the patient's lesion was more likely to occur within the CLSM maps. The nonparametric test results indicate that the mapping scores of the group with AF newly detected on ECG or cardiac monitoring after stroke are significantly higher than those of the group without AF newly detected on ECG or cardiac monitoring after stroke across FDC and SDC maps (Figure 5B). This suggests that the lesions in the group with AF newly detected on ECG or cardiac monitoring after stroke are more likely to fall within the proposed map regions. The logistic regression model, accounting for age, sex, NIHSS score, lesion volume, and cardiogenic factors as covariates, shows that the mapping scores of FDC and SDC have P values <0.05, with FDC scores exhibiting an odd ratio (OR) value of 1.27 (95% CI, 1.07–1.50) and SDC scores an OR of 1.43 (95% CI, 1.22–1.67) (Table 3).

Figure 5. The calculation method for the degree of overlap between each lesion image and the FDC map and SDC map (A) and violin plots of the mapping scores (B).

The P value represents the result of the nonparametric comparison. AFDAS indicates atrial fibrillation detected after stroke; FDC, functional disconnection connectome; and SDC, structural disconnection connectome.

Table 3.

Logistic Regression Results of the Mapping Scores

	β	t‐value	P value	OR	95% CI
Functional disconnection connectome score	0.24	2.79	0.005	1.27	1.07–1.50
Structural disconnection connectome score	0.36	4.49	<0.001	1.43	1.22–1.67
Sex	0.17	0.83	0.40	1.19	0.79–1.79
Age	0.06	6.21	<0.001	1.06	1.04–1.08
National Institutes of Health Stroke Scale score	0.07	2.30	0.003	1.07	1.02–1.12
Lesion volume	−0.01	−0.78	0.44	0.99	0.96–1.02
Left atrial diameter	0.18	10.67	<0.001	1.20	1.16–1.24
Left ventricular ejection fraction	−0.03	−2.09	0.04	0.97	0.95–1.00
Regional wall motion abnormality	0.20	0.62	0.54	1.22	0.65–2.26

OR indicates odds ratio.

DISCUSSION

This study provided the comprehensive map of network localizations for AF newly detected on ECG or cardiac monitoring after stroke. We discovered that AF newly detected on ECG or cardiac monitoring after stroke was correlated with impairments of brain network especially the disconnection in the CAN such as bilateral insula, temporal cortex, frontal cortex, and their related anatomical structures through SDC, FDC, as well as ROI analysis. The association remained robust even after adjustment for confounding variables.

In recent years, a series of studies applied lesion symptom mapping techniques to explore the associations between the infarct locations and cardiovascular complications. Previous studies have found that the lesion of the right dorsal anterior insular cortex can lead to the increase of troponin and the lesions of right insula, right parietal cortex, right frontal lobe, right amygdala, basal ganglia, and thalamus were significantly correlated with the occurrence of arrhythmia after acute ischemic stroke. ⁵ , ¹⁴ Another VLSM analysis demonstrated the associations between poststroke impaired left ventricular ejection fraction and lesions most prominently in the right posterior insular, amygdala, and adjacent opercular region. ¹⁵ Inconsistent with other studies, Rizos et al. failed to find specific brain areas that are associated with new AF after controlling for infarct size. The finding challenged the neurogenic hypothesis of AF that acute ischemic stroke in distinct brain regions of the central autonomic network may directly impair cardiac function. However, considering the central autonomic nervous system acting through sympathetic and parasympathetic networks plays a pivotal role in conveying central autonomic commands from the brain to the heart, it is not enough to analyze the lesion location instead of the network as a whole. The SDC and FDC combined with the sympathetic and parasympathetic network nodes analysis provided more comprehensive information on brain network disconnections induced by lesions.

Consistent with previous studies, we also found the lesions in the insular and disconnection of CAN are related to the occurrence of AF newly detected on ECG or cardiac monitoring after stroke. The insula is the control center of autonomic nerves ¹⁶ and is extensively connected to several sites that regulate autonomic nerves including the frontal lobe, amygdala, hypothalamus, and solitary nucleus. Balint et al. found that either left or right insular infarction of rats resulted in increased endothelial dysfunction, inflammation, and fibrosis in the left atrium at 28 days after stroke. ¹⁷ A recent study has revealed that persistent proinflammatory changes in monocytes/macrophages in multiple organs notably in the heart can lead to cardiac fibrosis and dysfunction in both mice and patients with stroke up to 3 months after brain injury. ¹⁸ According to a study exploring the associations between cardiovascular autonomic dysfunction and multiple sclerosis‐related lesion sites, left insular and hippocampal lesions were associated with an increase in sympathetic cardiovascular modulation. ¹⁹ These findings suggest a specific role of insula lesions in the pathogenesis of sympathetic hyperactivation and immunosuppression induced by stroke. ²⁰ Overactivation of the sympathetic nerve and decreased parasympathetic nerve activity caused by the damage of the central autonomic neural network as well as a state of high inflammation are the main pathogenesis of brain‐heart syndrome. ² , ²¹ , ²²

Besides, we also identified the role of the structural and functional disconnection of the frontal cortex and its related anatomical structures or fibers such as temporal lobe, amygdala, parahippocampal gyrus, fronto‐occipital fasciculus, and inferior longitudinal fasciculus. Previous studies have shown that unilateral lesions of the ventromedial prefrontal cortex have side‐dependent effects on heart rate and blood pressure responses to visual emotional stimuli. ²³ Sakaki et al. also found that greater heart rate variability was linked with stronger functional connectivity between amygdala and ventrolateral prefrontal cortex in younger than in older adults. ²⁴ Recent years a series of meta‐analyses suggests promising evidence that noninvasive brain stimulation serves as an effective tool for influencing cardiovascular functions. Noninvasive brain stimulation produced the largest effects on both measures of vagally mediated heart rate variability (ie, high frequency‐heart rate variability and RMSSD) when applied over the dorsolateral prefrontal cortex. ²⁵

This study, through an innovative multimodal neuroimaging analysis system, systematically elucidated the association between the network characteristics of infarction lesions and AF newly detected on ECG or cardiac monitoring after stroke. Compared with previous AF prediction models based on clinical variables and biomarkers, ²⁶ this study had the following advantages: First, this study revealed the association between brain network damage and AF newly detected on ECG or cardiac monitoring after stroke through comprehensive lesion network mapping analysis. By employing voxel‐based lesion‐symptom mapping and connectomics analysis at the whole‐brain voxel level, we revealed a network‐level damage pattern of AF newly detected on ECG or cardiac monitoring after stroke. The structural and functional disconnection in the frontal cortex‐insula‐amygdala pathway collectively constitute the neuroimaging signature of AF newly detected on ECG or cardiac monitoring after stroke. This multiscale analytical approach overcomes the limitations of traditional VLSM studies. Second, besides the traditional brain network, we explored the association between the central autonomic network and the occurrence of AF newly detected on ECG or cardiac monitoring after stroke for the first time. Our finding provided evidence for the neurogenic factors involving autonomic mechanisms that related to stroke‐induced heart injury. Most important, the network disconnection phenomena observed in patients with AF newly detected on ECG or cardiac monitoring after stroke may offer potential targets for neuromodulation therapy. Future clinical trials could explore whether noninvasive neuromodulation techniques, such as transcranial magnetic stimulation, targeting these regions can affect the incidence of stroke‐heart syndrome, especially AF newly detected on ECG or cardiac monitoring after stroke, in patients with stroke.

This study still has some limitations. First, we cannot definitely rule out that patients had undetected AF before the study. Hence, our results demonstrate associations between AF newly detected on ECG or cardiac monitoring after stroke and ischemic stroke lesions and networks, yet we cannot conclude that AF newly detected on ECG or cardiac monitoring after stroke is a result of the stroke. Second, we cannot define whether autonomic tone mediated the observations as no specific measure of autonomic function was performed in our study. Third, our findings should be interpreted considering the heterogeneous distribution of stroke causes in the study population. Notably, only 12‐lead ECG and 24‐hour Holter monitoring were performed in this study; thus, the number of patients with AF newly detected on ECG or cardiac monitoring after stroke may be underestimated. This also partly explained the relatively lower proportion of cardioembolism in our population, which could have weakly biased statistical comparisons of AF newly detected on ECG or cardiac monitoring after stroke and sinus rhythm. Besides, as lesion locations and patterns on diffusion‐weighted imaging are associated with stroke subtypes, ²⁷ pathogenetic distribution may influence the results of lesion mapping and connectomes. It should be cautious when interpreting our findings in populations with different distribution of stroke subtypes. Fourth, a total of 1235 patients were excluded due to incomplete cardiac monitoring data. Although our systematic comparison revealed minor differences in NIHSS scores and body mass index between the included and excluded groups (with all other baseline characteristics remaining balanced), this exclusion may still have a slight impact on the generalizability of our finding. Lastly, the mean infarct volume of the patients in this study was relatively small, and larger studies are still needed to analyze the association between infarct site and AF newly detected on ECG or cardiac monitoring after stroke in patients with larger infarct volumes in the future.

CONCLUSIONS

In summary, we found the significant structural and functional disconnection for patients with AF newly detected on ECG or cardiac monitoring after stroke and verified the association between the impairment of CAN and AF newly detected on ECG or cardiac monitoring after stroke. This study provides deep insights into the neuroanatomical basis of AF newly detected on ECG or cardiac monitoring after stroke and may offer specific local targets for future neuromodulation therapies.

Sources of Funding

This work was supported by grants from the National Natural Science Foundation of China (82001237), Beijing Hospitals Authority Youth Programme (QML20230503) and Postdoctoral Fellowship Program of China Postdoctoral Science Foundation (GZC20252760).

Disclosures

None.

Supporting information

Tables S1–S2

Figures S1–S2