Abnormal topological structure of structural covariance networks based on fractal dimension in noise induced hearing loss

Abstract

The topological attributes of structural covariance networks (SCNs) based on fractal dimension (FD) and changes in brain network connectivity were investigated using graph theory and network-based statistics (NBS) in patients with noise-induced hearing loss (NIHL). High-resolution 3D T1 images of 40 patients with NIHL and 38 healthy controls (HCs) were analyzed. FD-based Pearson correlation coefficients were calculated and converted to Fisher’s Z to construct the SCNs. Topological attributes and network hubs were calculated using the graph theory. Topological measures between groups were compared using nonparametric permutation tests. Abnormal connection networks were identified using NBS analysis. The NIHL group showed a significantly increased normalized clustering coefficient, normalized characteristic path length, and decreased nodal efficiency of the right medial orbitofrontal gyrus. Additionally, the network hubs based on betweenness centrality and degree centrality were both the right transverse temporal gyrus and left parahippocampal gyrus in the NIHL group. The NBS analysis revealed two subnetworks with abnormal connections. The subnetwork with enhanced connections was mainly distributed in the default mode, frontoparietal, dorsal attention, and somatomotor networks, whereas the subnetwork with reduced connections was mainly distributed in the limbic, visual, default mode, and auditory networks. These findings demonstrate the abnormal topological structure of FD-based SCNs in patients with NIHL, which may contribute to understand the complex mechanisms of brain damage at the network level, providing a new theoretical basis for neuropathological mechanisms.

Introduction

Noise-induced hearing loss (NIHL) is sensorineural hearing loss caused by long-term exposure to occupational, environmental and other loud sound sources¹. The World Health Organization has estimated that one-third of hearing loss could be attributed to noise exposure, and hearing loss has become a global public health problem^2,3. NIHL is permanent and irreversible in most cases, which causes many damages to the auditory structure and contributes to a great negative impact on the quality of life of individuals, such as depression, anxiety, and cognitive disorders^1,4. Given its widespread impact, it is crucial to understand the neuropathological mechanisms underlying NIHL. Owing to its non-invasiveness and high resolution, magnetic resonance technology has been widely used in clinics and has become a powerful tool for probing the human brain neuroscience. The researchers found that hearing loss was associated not only with auditory brain regions, but also with structural and functional changes in other regions. The structural changes were found in children with congenital bilateral deep SNHL after hearing deprivation⁵. Functional networks altered in infants with severe bilateral congenital sensorineural hearing loss⁶. Zou et al. found that the topological structure of the white matter network was disarranged in patients with unilateral sudden deafness, and the changes of brain node centers in SSNHL patients involved auditory network, visual network, attention network, default mode network, sensorimotor network and subcortical network⁷. These findings emphasize the necessity of comprehensively exploring the changes in brain structure and function in patients with NIHL. To explore the impact of NIHL on the brain more deeply and comprehensively, this study comprehensively uses multiple methods, such as fractal dimension (FD), graph theory, and network-based statistics (NBS) to study the changes in structural covariant networks (SCNs) in patients with NIHL and further explore the neuropathological mechanism of NIHL.

As a nonlinear measure of complexity, FD has unique advantages in evaluating the complexity of brain structure and can provide an important perspective for our research. FD is a nonlinear metric based on fractal geometry and has been defined as the complexity index for a long time. It summarizes the morphological characteristics of objects with providing the value of self-similarity, and then provides the value of its overall complexity⁸. FD has been used to evaluate the complexity of multiple scales on the brain^9,10. Studies have shown that in various psychiatric and nervous system diseases, FD is more sensitive to detect morphological changes related to pathological and developmental changes, and also can more sensitively characterize structural differences of brain^11–13. In the study of NIHL, FD also shows unique value. Compared with other methods that simply measure the size or morphology of the brain structure, FD focuses more on the complexity and self-similarity of the structure. It can quantify the subtle and complex changes that occur in the microstructure of brain, which are difficult to detect by traditional methods.

In addition, SCNs also have an important role in studying the impact of NIHL on the brain. SCNs have been applied to the study of various central nervous system diseases, such as schizophrenia, anxiety, and depression^14–16. Neuroimaging studies have shown that the theoretical basis of SCNs model is the common morphological changes of interconnected brain regions, which is a valuable tool for the study of brain topology¹⁷. SCNs can explore how different brain regions throughout the brain interact and affect their structural morphology, rather than just focusing on individual regions or structures^18,19. Compared with functional networks, SCNs have lower computational intensity and noise sensitivity, therefor, they have stronger persuasiveness and wider application in constructing group-level networks²⁰. Since NIHL may trigger synergistic changes among multiple brain regions, these changes may not be limited to auditory-related regions, but may also involve other regions related to functions such as cognition and emotion. Through SCNs analysis, it is possible to better understand how NIHL affects the structural covariant relationships among brain regions and then reveal the influence mechanism on overall brain function.

In order to better understand structural covariant relationship and its effect on brain function, researchers introduced graph theory to analyze brain networks^14,21,22. A complex network is formed in structure and function on the human brain, which can promote the effective processing and integration of information. Graph theory provides a powerful tool to model brain tissue, building a large-scale network of nodes and edges. Graph theory-based complex network analysis provides a powerful and general framework, which can directly measure the global and local organizational properties of networks, while also provides a range of metrics²³. Graph theory can help quantitatively analyze changes in the topological properties of the brain network in patients with NIHL. By comparing the topological differences of the brain network between NIHL patients and the normal population, we can deeply understand the impact of NIHL on the integration and separation of overall brain function. NBS analysis is used to solve multiple comparison problems when analyzing connectivity graphs and allowed to identify subnetworks and connections with abnormal connections. It can also reveal changes in the strength of connections in brain networks and provides a complementary approach for assessing interbrain connectivity differences²⁴. Since the brain network connections are complex and there may be multiple potentially abnormal connections in patients with NIHL. When processing such complex network data, traditional statistical methods may lead to a high false positive rate due to multiple comparison problems, whereas NBS can more accurately locate abnormal connections in brain networks. Graph theory and the NBS method have been widely used to treat hearing loss^5,6,12.

This study conducted a complex network analysis by comprehensively employing FD, graph theory, and NBS methods. It not only focused on regional changes, but also considered the overall interaction of the brain network, revealing the specific impacts of NIHL on the brain network structure. This study solved the deficiencies of existing studies on the changes in brain network structure and function in patients with NIHL, but also deeply explored the neural mechanism of NIHL from micro to macro and from local to global dimensions. This provides a new theoretical basis for revealing neuropathological mechanisms in patients with NIHL.

Materials and methods

Subjects

According to the national occupational health standard GBZ 49-2014 “Diagnosis of Occupational Noise Deafness” of the People’s Republic of China, 40 patients with NIHL diagnosed in the occupational department of Yantaishan Hospital from 2014 to 2020 were selected. In the same period, 38 healthy controls (HCs) with matching age and education level were selected. All subjects were tested using the Mini-Mental State Examination (MMSE) and Hamilton Anxiety Scale (HAMA). The inclusion criteria for the NIHL group were as follows: adult male; han nationality; education level from primary school to university level; normal mental state (MMSE ≥ 27), no other neurological, psychiatric history and family history; no history of psychotropic drugs. Exclusion criteria: illiterate or non-cooperative, or MRI scan contraindications.

The study protocol was approved by the Hospital Ethics Committee (Yanshanlun 2023014). Written informed consent was obtained from each subject. All methods were performed in accordance with the relevant guidelines and regulations.

MRI data acquisition and data preprocessing

MRI data acquisition

The subjects were scanned using GE Discovery MR 750 3.0T (GE Company, USA). MRI parameters were as follows: repetition time (TR) = 6.9 ms, echo time (TE) = 3.4 ms, flip angle = 12°, matrix size = 256 × 256, field of view (FOV) = 25.6 cm × 25.6 cm, scanning time = 258 s, 172 sagittal slices, a thickness of 1 mm without a gap. During the scanning process, the subjects maintained a natural supine position, closed their eyes, relaxed and reduced mental activity and head movement.

Data preprocessing

Data preprocessing was done in the software package SPM12 (https://www.fil.ion.ucl.ac.uk/spm/) and its CAT12 toolkit (https://neuro-jena.github.io/cat/). First, the original images in DICOM format were converted to NIFTI format, and all T1 images were manually checked for quality by the researchers. Secondly, the 3D-MRI images were segmented by the segment function of CAT12 software to obtain gray matter, white matter and cerebrospinal fluid images. Next, the Dartel algorithm was used to standardize the segmented images into the Montreal Neurological Institute (MNI) standard space to eliminate individual differences in the brain. Subsequently, segmented images were smoothed using a 20 mm full-width-at-half-maximum (FWHM) Gaussian kernel^13,25 to reduce the image registration error and improve the signal to noise ratio and image quality. Finally, the FD values of 68 brain regions were extracted using the CAT12 toolkit based on the Desikan-Killiany map²⁶, and were found to be normally distributed according to the Shapiro-Wilk test.

Construction of structural covariance networks

The FD values with 68 cortical regions were used to construct the SCNs. The 68 brain regions (ROIs) were defined as nodes of SCNs, and the Pearson correlation coefficients were defined as edges of SCNs. With age and education level as covariates, 68 × 68 correlation matrixes were established by calculating the Pearson correlation coefficients between FD values of each brain regions for each group^22,27.

In order to enhance the normality of the correlation coefficients, the values of 68 × 68 correlation matrix were transformed with Fisher’s Z. Subsequently, the correlation matrixes after Fisher’s Z were transformed to binary adjacency matrixes using various sparsity thresholds. Considering the absolute threshold at a single sparsity resulting in different numbers of edges, so the sparsity thresholds were defined as a range of network sparsity, ranging from 0.06 to 0.3 with an interval of 0.01. In this study, the minimum sparsity should meet the requirement that both SCNs were fully connected without any isolated nodes, so the minimum sparsity is 0.06^19,28. Simultaneously, 100 random networks were generated and used to calculate network properties to determine whether the SCNs of the two groups had significant non-random topology in the above process.

Graph-based network analysis

Network topology attributes

Network topology attributes of the SCNs were calculated using Brain Connectivity Toolbox (BCT, https://www.nitrc.org/projects/bct). Global network topology attributes include clustering coefficient (Cp), characteristic path length (Lp), normalized clustering coefficient (γ), normalized characteristic path length (λ), small-worldness (σ), local efficiency (Elocal) and global efficiency (Eglobal). Nodal topology attributes include betweenness centrality (BC), degree centrality (DC) and nodal efficiency (NE). The definitions of these network topology attributes were the same as in previous studies^21,29. The γ (γ = Cp-real / Cp-rand) and λ (λ = Lp-real / Lp-rand) were obtained by the ratio of Cp and Lp of 100 random networks to the real network respectively. The σ (σ = γ / λ) was obtained by the ratio of γ and λ.

Network hubs

The network hubs are the most interconnected region in the whole brain, which play a key role in the transmission and exchange of network information. The network hubs were defined as regions where the values of BC or DC were at least two standard deviations greater than the mean values^22,29.

Statistical analysis

Baseline characteristics

Demographic and clinical data were analyzed with SPSS 26 software. An independent two-sample t-test was performed to analyze differences in age, education level, HAMA score and MMSE score between NIHL and HCs groups (p < 0.05 was statistically significant).

Network topology attributes statistic

The nonparametric permutation test is a statistical method that does not depend on the assumption of the data distribution. Given the typically unknown distribution of network topology attribute values, we used the non-parametric permutation test to analyze between-group differences in all network topology attributes between NIHL and HCs groups^19,29,30. Initially, the real network topology attributes were calculated separately for NIHL and HCs groups within network sparsity range (0.06–0.3, with an interval of 0.01). Next, an independent permutation was performed for each sparsity level. The FD values of all subjects were randomly reallocated to NIHL and HCs groups with the same number of samples as the original groups, and the network topology attributes were recalculated within the same network sparsity range. This process was repeated 1000 times to obtain the permutation distribution. In each permutation test, the differences in network topology properties between NIHL and HCs groups were calculated and a null distribution of these differences was generated, which reflected the range of differences that could occur under the null hypothesis. Subsequently, the actual between-group differences were placed into the null distribution generated above, and p-values were calculated based on the position of the actual differences in the percentile of the null distribution, thereby assessing the statistical significance of the differences. Within the sparsity range, significant differences were observed in a consistent direction. Significant differences were confirmed when between-group differences were less than 2.5% or greater than 97.5%. Furthermore, because the area under the curve (AUC) is an integrated scalar of indicators under a group of thresholds, it is not affected by the randomness and bias of a single threshold selection, and is more sensitive to subtle and complex changes in brain function and structure caused by diseases³¹. Therefore, this study used AUC analysis to quantify the topological attributes of each network. Considering that evaluating network topological attributes at different network sparsity will produce multiple comparisons, we used the false discovery rate (FDR) method to correct the AUC differences to improve the reliability of statistical inference. Finally, the AUC differences of network topology attributes in real SCNs were compared with those in random SCNs (1000 repetitions), and the FDR method was applied for multiple comparisons with a threshold of 0.05.

Network-based statistic

The NBS is a nonparametric statistics approach to identify abnormal subnetworks of altered structural connectivity in the brain²⁴. First, setting the threshold to 3.1 (t-test statistic threshold, correspond to p < 0.001), all edges were globally searched to identify any connected subnetworks of between-group differences with exceeding threshold value. Secondly, corrected p values for each subnetwork previously identified were calculated using permutation tests. P < 0.01 was set as the threshold of significance analysis. The null distribution for the largest subnetwork was generated under the null hypothesis of 1000 permutations, and then the statistical significances were evaluated. The subnetworks with increased and decreased connections were obtained by respective calculating NIHL-HCs and HCs-NIHL through the above process. Finally, the significant subnetworks were reported after corrected at p < 0.001, and were visualized using BrainNet Viewer³² and circos (https://circos.ca/).

In addition, based on previous research³³, the brain regions according to the Desikan–Killiany atlas were divided into eight networks, which were showed in Table 1.

Table 1.

Abbreviations of networks.

Network	Abbreviation	Network	Abbreviation
Visual network	VSN	Dorsal attention network	DAN
Somatomotor network	SMN	Frontoparietal network	FPN
Auditory network	AUN	Default mode network	DMN
Ventral attention network	VAN	Limbic network	LBN

Results

Demographic and clinical characteristics and FD values

The demographic and clinical characteristics of NIHL and HCs groups are shown in Table 2. There were no significantly differences in age (p = 0.930), education level (p = 0.281), MMSE (p = 0.253) between NIHL and HCs groups. The HAMA (p < 0.001) in NIHL group was significantly greater than HCs group. FD values in 68 brain regions of NIHL and HCs groups are shown in Table 3.

Table 2.

Demographic and clinical characteristics of all participants (mean ± SD).

	NIHL(n = 40)	HCs(n = 38)	t	p
Age(years)	45.50 ± 8.17	45.66 ± 7.74	− 0.088	0.930
Education level(years)	11.20 ± 1.86	11.74 ± 2.48	− 1.086	0.281
HAMA score	12.79 ± 12.32	3.76 ± 1.05	4.53	< 0.001*
MMSE score	29.15 ± 1.37	28.82 ± 1.18	1.15	0.253

NIHL noise-induced hearing loss, HCs healthy controls, HAMA Hamilton Anxiety Scale, MMSE mini-mental state examination.

*Represented significant differences (p < 0.05).

Table 3.

FD values of all participants (mean ± SD).

No.	Regions	Hemisphere	NIHL(n = 40)	HCs(n = 38)	t	p
1	Bankssts	L	2.90 ± 0.23	2.87 ± 0.20	0.713	0.478
		R	2.84 ± 0.15	2.87 ± 0.14	− 0.928	0.356
2	Caudal anterior cingulate gyrus	L	2.54 ± 0.14	2.54 ± 0.17	0.059	0.953
		R	2.43 ± 0.15	2.46 ± 0.14	− 0.761	0.449
3	Caudal middle frontal gyrus	L	2.81 ± 0.12	2.85 ± 0.12	− 1.188	0.239
		R	2.92 ± 0.14	2.91 ± 0.14	0.382	0.703
4	Cuneus	L	2.67 ± 0.16	2.69 ± 0.17	− 0.401	0.690
		R	2.71 ± 0.21	2.70 ± 0.16	0.172	0.864
5	Entorhinal cortex	L	2.61 ± 0.17	2.65 ± 0.19	− 0.966	0.337
		R	2.62 ± 0.17	2.61 ± 0.18	0.305	0.761
6	Fusiform gyrus	L	2.45 ± 0.09	2.45 ± 0.10	− 0.126	0.900
		R	2.43 ± 0.07	2.43 ± 0.08	− 0.281	0.779
7	Inferior parietal gyrus	L	2.79 ± 0.09	2.79 ± 0.10	0.080	0.937
		R	2.87 ± 0.10	2.89 ± 0.09	− 1.026	0.308
8	Inferior temporal gyrus	L	2.53 ± 0.14	2.52 ± 0.14	0.187	0.853
		R	2.50 ± 0.11	2.51 ± 0.09	− 0.467	0.642
9	Isthmus cingulate cortex	L	2.06 ± 0.11	2.08 ± 0.14	− 0.847	0.399
		R	2.04 ± 0.11	2.05 ± 0.11	− 0.497	0.621
10	Lateral occipital gyrus	L	2.50 ± 0.07	2.50 ± 0.07	2.610	0.011*
		R	2.42 ± 0.10	2.42 ± 0.08	− 0.395	0.694
11	Lateral orbitofrontal gyrus	L	2.28 ± 0.05	2.27 ± 0.06	0.918	0.361
		R	2.27 ± 0.07	2.27 ± 0.06	0.005	0.996
12	Lingual gyrus	L	2.64 ± 0.11	2.63 ± 0.11	0.236	0.814
		R	2.69 ± 0.09	2.67 ± 0.09	0.895	0.374
13	Medial orbitofrontal gyrus	L	2.60 ± 0.13	2.62 ± 0.10	− 1.085	0.281
		R	2.52 ± 0.12	2.54 ± 0.10	− 0.947	0.346
14	Middle temporal gyrus	L	2.62 ± 0.13	2.60 ± 0.12	0.619	0.538
		R	2.56 ± 0.11	2.59 ± 0.10	− 1.057	0.294
15	Parahippocampal gyrus	L	2.58 ± 0.16	2.60 ± 0.19	− 0.676	0.501
		R	2.46 ± 0.14	2.47 ± 0.13	− 0.449	0.654
16	Paracentral lobule	L	2.63 ± 0.12	2.63 ± 0.12	− 0.124	0.902
		R	2.60 ± 0.10	2.59 ± 0.10	0.505	0.615
17	Pars opercularis	L	2.83 ± 0.18	2.80 ± 0.19	0.733	0.466
		R	2.92 ± 0.18	2.87 ± 0.16	1.139	0.258
18	Pars orbitalis	L	2.78 ± 0.18	2.73 ± 0.15	1.242	0.218
		R	2.79 ± 0.15	2.85 ± 0.20	− 1.508	0.136
19	Pars triangularis	L	2.71 ± 0.13	2.69 ± 0.17	0.566	0.573
		R	2.69 ± 0.19	2.68 ± 0.18	0.366	0.716
20	Pericalcarine cortex	L	2.66 ± 0.22	2.71 ± 0.18	− 0.972	0.334
		R	2.40 ± 0.25	2.34 ± 0.22	1.148	0.254
21	Postcentral gyrus	L	2.80 ± 0.11	2.80 ± 0.09	0.135	0.893
		R	2.82 ± 0.10	2.79 ± 0.10	1.161	0.249
22	Posterior cingulate cortex	L	2.65 ± 0.16	2.65 ± 0.18	− 0.159	0.874
		R	2.52 ± 0.13	2.56 ± 0.15	− 1.034	0.304
23	Precentral gyrus	L	2.80 ± 0.07	2.79 ± 0.06	0.854	0.396
		R	2.77 ± 0.05	2.76 ± 0.07	0.685	0.496
24	Precuneus	L	2.74 ± 0.11	2.75 ± 0.10	− 0.652	0.517
		R	2.70 ± 0.09	2.67 ± 0.07	1.781	0.079
25	Rostral anterior cingulate gyrus	L	2.07 ± 0.12	2.09 ± 0.14	− 0.593	0.555
		R	2.07 ± 0.14	2.13 ± 0.14	− 1.637	0.106
26	Rostral middle frontal gyrus	L	2.53 ± 0.08	2.50 ± 0.08	2.029	0.046*
		R	2.51 ± 0.09	2.52 ± 0.09	− 0.653	0.516
27	Superior frontal gyrus	L	2.41 ± 0.05	2.42 ± 0.06	− 0.724	0.471
		R	2.38 ± 0.06	2.37 ± 0.04	0.227	0.821
28	Superior parietal gyrus	L	2.68 ± 0.07	2.66 ± 0.08	0.757	0.451
		R	2.73 ± 0.07	2.69 ± 0.08	1.786	0.078
29	Superior temporal gyrus	L	2.76 ± 0.12	2.77 ± 0.12	− 0.107	0.915
		R	2.74 ± 0.13	2.75 ± 0.10	− 0.326	0.745
30	Supramarginal gyrus	L	2.64 ± 0.09	2.68 ± 0.09	− 1.626	0.108
		R	2.62 ± 0.09	2.65 ± 0.08	− 1.449	0.151
31	Frontal pole	L	2.56 ± 0.18	2.60 ± 0.24	− 0.782	0.436
		R	2.57 ± 0.19	2.57 ± 0.19	− 0.024	0.981
32	Temporal pole	L	2.40 ± 0.13	2.45 ± 0.12	− 1.746	0.085
		R	2.35 ± 0.13	2.39 ± 0.12	− 1.399	0.166
33	Transverse temporal gyrus	L	2.48 ± 0.24	2.47 ± 0.23	0.231	0.818
		R	2.36 ± 0.18	2.41 ± 0.21	− 1.189	0.238
34	Insula	L	2.35 ± 0.07	2.39 ± 0.09	− 2.178	0.033*
		R	2.35 ± 0.09	2.36 ± 0.08	− 0.780	0.438

NIHL noise-induced hearing loss, HCs healthy controls, L left, R right.

*Represented significant differences (p < 0.05).

Global network topology attributes

The changes and between-group differences between NIHL and HCs groups in global network attributes at network sparsity range from 0.06 to 0.3 are shown in Fig. 1. Small-world properties were found in both NIHL and HCs groups, with γ > 1, λ ≈ 1 and σ > 1 in Fig. 1A, C,E. Compared with the HCs group, the NIHL group exhibited significantly increased λ and γ at several network densities in Fig. 1B, D. No significant between-group differences were found in σ, Cp, Elocal, Lp and Eglobal attributes in Fig. 1 (F, H, J, L, N).

Fig. 1.

Changes and between-group differences in global network attributes at different network sparsity (0.06–0.3) between NIHL and HCs groups. Graphs (A,C,E,G,I,K,M) showed changes, and graphs (B,D,F,H,J,L,N) showed between-group differences under different network sparsity in NIHL and HCs groups. The blue lines (-.) represented a 95% confidence interval, whereas the black lines (-) in the middle denoted the mean differences after 1000 permutations at each network sparsity. The pink lines (·) represented the real between-group differences between NIHL and HCs groups, which fell outside the confidence intervals and indicated significant between-group differences (p < 0.05) under the current threshold. The positive values indicated NIHL > HCs, and the negative values indicated NIHL < HCs. * represented significant differences.

AUC analysis showed that there are no significant differences between NIHL and HCs groups in global network attributes after FDR correction (p < 0.05).

Nodal topology attributes

Significant differences between NIHL and HCs groups were found in regional topological measures after FDR correction (p < 0.05) though AUC analysis. Compared with the HCs group, the NIHL group had decreased NE in the right medial orbitofrontal gyrus (MORG.R). However, no significant differences were found in BC and DC between two groups.

Network hubs analysis

Network hubs based on BC and DC in NIHL and HCs groups are shown in Fig. 2; Table 4. On both BC-based and DC-based network hubs, three hubs were found in the HCs group and two hubs in the NIHL group. The network hubs in the HCs group included left caudal anterior cingulate gyrus (CACG.L), left posterior cingulate cortex (PCC.L), right medial orbitofrontal gyrus (MORG.R) and right transverse temporal gyrus (TTG.R); while in the NIHL group included right transverse temporal gyrus (TTG.R) and left parahippocampal gyrus (PHG.L). In the NIHL group, the network hubs based on DC and BC were both the right transverse temporal gyrus (TTG.R) and the left parahippocampal gyrus (PHG.L), while in the HCs group, the left caudal anterior cingulate gyrus (CACG.L) and the right medial orbitofrontal gyrus (MORG.R) were the consistent network hubs.

Fig. 2.

Network hubs based on BC and DC in NIHL and HCs groups. Hubs based on BC were shown with green in HCs (A) and blue in HINL (B). Hubs based on DC were shown with pink in HCs (C) and yellow in HINL (D).The size of dot represented the values of BC and DC. NIHL noise-induced hearing loss, HCs healthy controls, BC betweenness centrality, DC degree centrality, L left, R right. Abbreviations were listed in Table 5.

Table 4.

Network hubs regions of HINL and HCs groups.

Hubs regions	BC	DC
HCs
CACG.L	89.77	9.18
PCC.L	85.51	/
MORG.R	81.20	8.86
TTG.R	/	9.47
NIHL
PHG.L	94.40	8.85
TTG.R	101.78	9.46

\NIHL noise-induced hearing loss, HCs healthy controls, BC betweenness centrality, DC degree centrality, L left, R right. Abbreviations were listed in Table 5.

Table 5.

Abbreviations of brain regions.

Regions	Abbreviation	Network	Regions	Abbreviation	Network
Bankssts	BSTS	DMN	Pars orbitalis	PORB	DMN
Caudal anterior cingulate gyrus	CACG	DMN	Pars triangularis	PTRI	FPN
Caudal middle frontal gyrus	CMFG	FPN	Pericalcarine cortex	PCAL	VSN
Cuneus	CUN	VSN	Postcentral gyrus	POSCG	SMN
Entorhinal cortex	ENC	LBN	Posterior cingulate cortex	PCC	DMN
Fusiform gyrus	FUG	VSN	Precentral gyrus	PRECG	SMN
Inferior parietal gyrus	IPG	FPN	Precuneus	PCUN	DMN
Inferior temporal gyrus	ITG	DAN	Rostral anterior cingulate gyrus	RACG	DMN
Isthmus cingulate cortex	ICC	DMN	Rostral middle frontal gyrus	RMFG	FPN
Lateral occipital gyrus	LOG	VSN	Superior frontal gyrus	SFG	DAN
Lateral orbitofrontal gyrus	LORG	DMN	Superior parietal gyrus	SPG	DAN
Lingual gyrus	LIG	VSN	Superior temporal gyrus	STG	AUN
Medial orbitofrontal gyrus	MORG	LBN	Supramarginal gyrus	SMG	VAN
Middle temporal gyrus	MTG	DMN	Frontal pole	FPO	DMN
Parahippocampal gyrus	PHG	LBN	Temporal pole	TPO	LBN
Paracentral lobule	PCL	SMN	Transverse temporal gyrus	TTG	AUN
Pars opercularis	POPE	FPN	Insula	INS	VSN

Abbreviations of network were listed in Table 1.

Network alterations based on NBS analysis

Two abnormal subnetworks were found according to NBS analysis results between NIHL and HCs groups, which were shown in Figs. 3 and 4; Table 6. In the NIHL group, the network of increased connections comprised 26 nodes and 29 connections, which were distributed throughout both hemispheres, in Fig. 3; Table 6. These alterations were primarily distributed in the network connectivity spanning the DMN, FPN, DAN and SMN. Inversely, the NIHL group exhibited network of decreased connections with 23 nodes and 24 connections. These reduced connections mainly involved VSN, LBN, DMN and AUN, in Fig. 4; Table 6.

Fig. 3.

Network of increased connections between NIHL and HCs groups based on NBS (p < 0.001). (A) Nodes and Edges of networks of increased connections. Colored dots represented nodes, and nodes were divided into different colors according to resting-state brain networks which were listed in Table 1. Lines represented edges, and the thickness of lines represented the strength of the connection. (B) The spatial distribution of networks of increased connections. The red lines represented edges with significantly increased connections. L left, R right. Abbreviations were listed in Table 5.

Fig. 4.

Network of decreased connections between NIHL and HCs groups based on NBS (p < 0.001). (A) Nodes and Edges of networks of decreased connections. Colored dots represented nodes, and nodes were divided into different colors according to resting-state brain networks which were listed in Table 1. Lines represented edges, and the thickness of lines represented the strength of the connection. (B) The spatial distribution of networks of decreased connections. The blue lines represented edges with significantly decreased connections. L: left, R: right. Abbreviations were listed in Table 5.

Table 6.

Nodes and edges with abnormal networks in NIHL group.

Network with increased connections (NIHL > HCs)			Network with decreased connections (NIHL < HCs)
Network-node		Network-node	Network-node		Network-node
DMN-BSTS.L	–	DAN-SPG.L	LBN-TPO.L	–	LBN-MORG.R
DMN-LORG.L	–	DAN-ITG.R	LBN-ENC.L	–	DMN-CACG.L
DMN-LORG.L	–	DMN-BSTS.R	LBN-MORG.R	–	DMN-CACG.L
DMN-CACG.L	–	FPN-CMFG.R	LBN-TPO.R	–	DMN-PCUN.L
DMN-CACG.L	–	FPN-IPG.R	LBN-PHG.L	–	DMN-ICC.L
DMN-MTG.R	–	SMN-PRECG.R	LBN-PHG.R	–	VSN-INS.R
DMN-BSTS.R	–	DAN-SPG.R	DMN-BSTS.L	–	AUN-STG.L
FPN-IPG.L	–	FPN-POPE.R	DMN-CACG.L	–	LBN-TPO.R
FPN-POPE.L	–	FPN-IPG.R	DMN-CACG.L	–	VSN-PCAL.R
FPN-PTRI.L	–	DMN-MTG.R	DMN-ICC.L	–	DMN-PCUN.L
FPN-RMFG.R	–	SMN-PRECG.R	DMN-PCUN.R	–	LBN-PHG.L
FPN-CMFG.R	–	DMN-PCC.L	VSN-INS.R	–	SMN-PCL.L
FPN-POPE.R	–	DAN-SPG.L	VSN-PCAL.R	–	AUN-TTG.R
FPN-IPG.R	–	AUN-TTG.R	VSN-LIG.L	–	AUN-TTG.R
DAN-SPG.L	–	FPN-PTRI.L	VSN-PCAL.L	–	AUN-TTG.R
DAN-SFG.L	–	FPN-IPG.R	VSN-LOG.R	–	VSN-PACL.L
DAN-SFG.L	–	DAN-SFG.R	AUN-STG.L	–	FPN-PTRI.L
DAN-SFG.R	–	FPN-IPG.R	AUN-STG.L	–	VSN-LIG.L
DAN-SFG.R	–	SMN-PRECG.R	AUN-TTG.R	–	SMN-PCL.L
DAN-ITG.R	–	LBN-PHG.L	FPN-PTRI.L	–	LBN-TPO.R
DAN-SPG.R	–	DMN-CACG.L	FPN-CMFG.L	–	VSN-LIG.L
SMN-POSCG.L	–	AUN-TTG.R	FPN-CMFG.L	–	LBN-MORG.R
SMN-PRECG.L	–	DMN-MTG.R	SMN-PRECG.R	–	DMN-PCUN.R
SMN-PRECG.R	–	AUN-TTG.R	VAN-SMG.R	–	LBN-TPO.L
SMN-POSCG.R	–	AUN-TTG.R
LBN-PHG.L	–	FPN-PTRI.L
LBN-MORG.R	–	FPN-IPG.L
AUN-TTG.R	–	FPN-POPE.L
VAN-SMG.R	–	SMN-PRECG.R

NIHL noise-induced hearing loss, HCs healthy controls, L left, R right. Abbreviations were listed in Tables 1 and 5.

Discussion

In this study, we found that the topology characteristics and network connections of FD SCNs were changed in patients with NIHL. The specific performances were as follows: (1) at the global level, the SCNs of both NIHL and HCs groups exhibited small-world properties. Compared with the HCs group, the λ and γ were more increased in the NIHL group. (2) at the nodal level, the NE of some nodes in the NIHL group were lower than those of the HCs group. (3) on BC-based and DC-based network hubs, the centers were located in different regions in NIHL and HCs groups respectively. (4) two abnormally connected networks appeared in the NIHL group.

In general, these studies contribute to our understanding of the neuropathology of NIHL at the network level, and improve our knowledge of the effects of NIHL on brain structure.

Alterations in global attributes of FD SCNs in NIHL

Previous studies had showed that both the structural and functional networks of the human brain followed a small-world topology. Small-world characteristics reflected the best balance between local separation and integration, reflecting a balanced network state^34,35. Small-world networks supported local specialization and integrated processing through efficient information transfer while minimizing cabling costs^36,37. In this study, the SCNs of NIHL and HCs groups both had small-world characteristics. This suggests that a high level of information interaction in the brain network in patients with NIHL is still present. However, some topological attributes of the NIHL group had altered. Compared with the HCs group, the NIHL group showed significantly increased γ and λ. Both of the unilateral SSNHL groups (lSSNHL and rSSNHL group) showed a significantly increased normalized clustering coefficient (γ) and normalized characteristic path length (λ)³⁸. Wang et al. found that the normalized clustering coefficient and normalized path length increased in patients with mild cognitive impairment, and the abnormality of these structural networks was related to the slow information processing speed³⁹. These are consistent with our findings. The increase in λ usually means that the information transmission path in the brain network becomes more lengthy, resulting in a decrease in information transmission efficiency. In patients with NIHL, this change may be due to the fact that noise exposure leads to a reduction in brain network connections or a weakening of connection strength, so that the information needs to pass through more intermediate nodes during the transmission process, thereby increasing the transmission time and complexity. This indicates that the global information integration ability of the brain network declines in patients with NIHL, which may further affect the speed and accuracy of the information processing in the brain. Meanwhile, the increase in γ may reflect the enhancement of local connections in the brain network. This may be due to noise exposure, which causes the brain to form more tightly connected clusters in local areas in an attempt to maintain or restore local information processing capabilities. This local adaptive response may be a compensation mechanism for noise damage in the brain.

In normal human brain networks, higher levels of two parameters (Cp and Elocal), which measure functional separation of the network, indicate increased local information processing capacity in the brain. Cp refers to the degree of interconnection between network nodes and neighboring nodes, reflecting the tightness of the connection of the local brain network, and is used to evaluate the efficiency of local information processing in the brain network⁴⁰. Higher Cp reflects network isolation in brain networks, which is the ability to perform specialized neuronal processing between tightly connected regions⁴¹. Elocal is the average of the reverse shortest path lengths between adjacent nodes. It quantifies the efficiency of information transmission between nodes and neighboring nodes and is used to measure the degree of isolation of the network⁴². Compared with the HCs group, the NIHL group had a higher Cp and Elocal, although there was no significant difference. These also indicated that the information processing ability of local brain areas was enhanced and the local information transmission efficiency didn’t decrease significantly in patients with NIHL. This was consistent with previous studies of age-related hearing loss and unilateral sundden sensorineural hearing loss^5,38. The brains of patients with NIHL showed an increase in the clustering coefficient and stability in local efficiency, suggesting that the brain may optimize the efficiency and accuracy of information processing by enhancing the connections between specific brain regions. Simultaneously, the brain may aim to maintain or even improve the efficiency of information transmission by strengthening internal connections. Although the differences in Cp and Elocal between NIHL and HCs groups did not reach a statistically significant level, this trend still provides us with important clues: the brains of patients with NIHL may undergo an adaptive or compensatory adjustment process. This process enables the brain to maintain or enhance the information processing capabilities of local brain regions as well as the efficiency of information transfer, thus compensating for the functional loss caused by impaired auditory input to a certain extent.

This suggests that the increased separation of SCNs in patients with NIHL may be a compensatory mechanism for brain structural damage. In a word, the topological properties of FD SCNs have changed in patients with NIHL. This may provide further evidence that NIHL is a disease associated with brain network alterations.

Alterations in nodal attributes of FD SCNs in NIHL

The abnormal topological properties of nodes indicate that the connectivity and communication efficiency of nodes with other regions are changed, resulting in abnormal regional neural circuits of brain networks, which can obtain additional information that can’t be obtained from global topological networks⁴³. BC refers to the number of shortest paths between all pairs of nodes in the network through a node, which measures the influence of nodes on the overall information flow in the brain network⁴⁴. DC indicates the number of edges directly connected to the node. The DC is greater, the nodes are connected more. DC evaluates the importance of nodes in the network⁴². NE indicates the information communication capability of a node in a network³⁵.

At the level of nodes, compared with the HCs group, the node with significant difference in NE was found in the NIHL group, which was MORG.R. This abnormal node was also found in BC and DC, but it was not corrected by FDR (p < 0.05). MORG is involved in higher cognitive processes within the reward circuit that is necessary for reward processing, suppression of emotional responses, and habit formation, and it promotes behavioral outcomes⁴⁵. Previously, abnormalities in MORG nodes were also found in some anxiety disorders, mental disorders, depression and other diseases. For example: abnormalities of the medial orbitofrontal cortex had been observed in various anxiety disorders, and it was believed that the medial orbitofrontal cortex was involved in the processing of trait anxiety⁴⁶, functional connections between the right medial orbital gyrus and the dopamine region of the brain are reduced in patients with first-episode schizophrenia⁴⁷, medial orbitofrontal cortex voxels had lower functional connectivity with temporal cortex areas, the parahippocampal gyrus, and fusiform gyrus in depression⁴⁸.

In patients with NIHL, changes in the efficiency of MOFG nodes may be due to the remodeling of brain structure and function triggered by long-term noise exposure. The abnormal nodes may mean that some advanced functional interactions of SCNs in patients with NIHL are interrupted, resulting in reduced connectivity and communication efficiency between nodes and other brain areas.

Alterations in nodal hubs

In brain networks, central nodes are at the intersection of many short paths that control the flow of information through the brain⁴². Central nodes represent areas of the brain that are highly connected and highly centered, which are very important in brain networks but are susceptible to disease-related processes⁴⁹. The changes of the center are closely related to neurological and psychological diseases^50,51.

In this study, the distributions and numbers of BC-based and DC-based hubs were found to be different in the NIHL group than the HCs group. The patients with NIHL had the same nodal hubs based on BC and DC, including the PHG.L and the TTG.R. The PHG belongs to limbic network which is mainly involved in situational cognition, emotion, memory and attention. The limbic network is considered to be able to directly or indirectly accept the input of the auditory signal and produce the corresponding effect⁵². Hua et al.⁵³ found that the centrality of the parahippocampal gyrus increased in sudden sensory-hearing loss. Moreover, the TTG belongs to the auditory network and is closely related to auditory information processing⁵⁴. Wen’s study found that the apparent diffusion coefficient value of the temporal transverse gyrus increased⁵⁵. Elevated ALFF in the temporal transverse gyrus was found in patients with low frequency amplitude senile deafness⁵⁶. Zhang found that the core nodes of the right deafness group included the parahippocampal gyrus and the superior temporal gyrus related to hearing⁵⁷. Feng et al. found that ALFF in the transverse temporal gyrus and parahippocampal gyrus in patients with unilateral sensorineural hearing loss were significantly higher than the control group⁵⁸. These are similar to our findings. In the NIHL group, the PHG.L and TTG.R served as nodal centers, while in the HCs group, the nodal centers were CACG.L and MORG.R. The changes in nodal hubs in the brains of patients with NIHL may be due to brain plasticity and adaptive adjustments caused by long-term noise exposure. The PHG.L and TTG.R, serving as nodal hubs, may be to adapt to the challenges brought by hearing loss and redistribute brain resources to process auditory and other related information more effectively. These obvious changes in nodal hubs suggest that the brain network has been reorganized laterally after NIHL, and the functions of other sensory channels have been compensated and enhanced.

Abnormal network based on NBS

In the NIHL group, the subnetworks with enhanced connections were mainly distributed in DMN, FPN, DAN and SMN. The increased connections between nodes mainly occurred between DMN-DAN, DMN-FPN and DAN-FPN. The DMN is a network that is closely related to the integration of cognitive and emotional processing and is responsible for directing thinking and mental activity⁵⁹. Li et al. found that patients with SSNHL had increased network connectivity between the DAN, FPN and DMN³³. In elderly patients with moderate hearing loss, the auditory cortex showed activation in response to somatosensory stimulus⁶⁰. Reduced auditory activity led to increased somatosensory activity⁶¹. A study of unilateral hearing loss in children found co-activation between multiple FPN and DMN regions, and increased connectivity between DMN and FPN⁶². The study suggested that the response to sensory deficits was achieved through dynamic reorganization of internetwork connections, possibly reflecting compensatory adaptations to maintain cognitive and sensory functions^63,64. Therefore, after the auditory channel is damaged in patients with HINL, the function of other sensory channels is also affected. The compensation of auditory perception may be realized through cross-channel remodeling.

Conversely, the subnetworks with reduced connectivity in the NIHL group were mainly distributed in LBN, VSN, DMN and AUN. The reduced connections between nodes mainly existed between AUN-VSN and LBN-DMN. Ranran et al. found that patients with severe NIHL had reduced functional network connectivity within the AUN, VSN and DAN in the independent component analysis study⁶⁵. A study using visuospatial working memory tasks found that patients with single-sided deafness had reduced activation in both auditory and visual cortices⁶⁶. The node connectivity of the default network, visual network, and subcortical network was significantly reduced in patients with long-term unilateral hearing loss⁶⁷. In patients with SSNHL, connections between VSN, AUN and DAN showed a significant decrease in Internet connection strength³³. These results show that NIHL causes significant damage to auditory and sensory cognitive memory related networks. These deficiencies suggest that the effects of NIHL on brain tissue function are complex, and its potential ripple effects may extend to various cognitive and sensory functions.

It is worth noting that we found 41 node connections changed (excluding duplicate nodes), and these nodes were distributed across eight networks. Li et al. found significant between-group differences in the overall functional connectivity of 55 nodes within nine networks in patients with SSNHL, and these changes highlighted the widespread effects of SSNHL on functional connectivity on the brain³³. The changes of brain nodes in patients with SSNHL involved auditory network, visual network, attention network, default mode network, sensorimotor network and subcortical network, and the WM network topology changed to randomization⁷. These changes indicate that there are extensive changes in the internal connectivity of nodes after NIHL, with a larger range of effects than traditional auditory networks.

The abnormal subnetworks suggest that the loss of function coexists with compensatory network recombination in the SCNs in patients with NIHL, suggesting that NIHL is not only a hearing disorder, but also a central nervous system disease.

There are some limitations in this study. First, the SCNs are built on the basis of group level, and each subject doesn’t have a separate network, so the correlation between network parameters and clinical indicators can’t be discussed. Second, the sample size is small, and the results may have statistical bias. Third, this study is a cross-sectional study and cannot reflect the dynamic changes of SCNs in patients with NIHL. In the future, it is necessary to continue to expand the sample size and conduct longitudinal studies to further clarify the relationships between SCNs and functional activity and cognition in patients with HINL.

Conclusion

In this study, we used graph theory, NBS and FD to study changes of SCNs in patients with NIHL. We found that SCNs in patients with NIHL had small-world characteristics, but had abnormal topological properties. There were different network hubs in NIHL and HCs groups. In addition, based on NBS analysis, two abnormal subnetworks were found with extensive changes in patients with NIHL. These findings may contribute to understand the complex underlying mechanisms of brain damage at the network level in patients with NIHL, and will provide new theoretical basis for the neuropathological mechanisms.

Author contributions

Wang Liping: Conceptualization, Investigation, Methodology, Project Management, Writing-original draft, Software. Lv Minghui: Data curation, Methodology, Investigation, Writing-review & editing. Zhang Jiayuan: Investigation, Software, Visualization. Wai Aijie: Data curation, Methodology, Investigation, Visualization. Huang Ranran: Conceptualization, Methodology, Writing-review & editing. Zhang zengcai: Conceptualization, Methodology, Software, Writing-review & editing, Supervision.Zhang guowei: Conceptualization, Funding acquisition, Resources, Writing-review & editing, Supervision.

Data availability

Due to privacy or ethical restrictions, the data of this study are not publicly available. But data are available upon reasonable request from the corresponding author.

Declarations

Competing interests

The authors declare no competing interests.

Ethical statement

The study protocol was approved by the Hospital Ethics Committee (Yanshanlun 2023014). Written informed consent was obtained from each subject. All methods were performed in accordance with the relevant guidelines and regulations.