The micro-structural changes in white matter fibers associated with anxiety and depression in moderate-severe obstructive sleep apnea

Abstract

This study compared the effectiveness of Automated Fiber Quantification (AFQ) and Tract-Based Spatial Statistics (TBSS) in detecting white matter (WM) changes in patients with moderate-to-severe obstructive sleep apnea (OSA), aiming to identify the more suitable method for investigating WM alterations of OSA patients. It also explored the potential neuropathological links between WM changes and anxiety/depression in OSA. Thirty-three untreated patients with moderate-severe OSA and 28 good sleepers without sleep disorders underwent 3.0T MRI scans. TBSS was used to construct WM skeletons and compare diffusion tensor imaging (DTI) metrics between groups. AFQ was then applied to extract 20 major fiber tracts from each subject, with each tract divided into 100 nodes to precisely localize affected areas. TBSS revealed no significant group differences after correction. In contrast, AFQ identified significant changes: decreased fractional anisotropy at nodes 77–78 of the forceps major (positively correlated with anxiety scores); increased mean diffusivity at nodes 23–24 of the right thalamic radiation; and increased axial diffusivity at nodes 54–57 of the right inferior fronto-occipital fasciculus (negatively correlated with anxiety and depression scores), and nodes 66–68 of the right superior longitudinal fasciculus. AFQ is more sensitive than TBSS in detecting focal WM changes in OSA, allowing precise localization of affected fiber nodes. These WM alterations, particularly in the corpus callosum and right inferior fronto-occipital fasciculus, are linked to anxiety and depression, offering insights into the neuropathological mechanisms of anxiety and depression in OSA.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-026-54091-1.

Introduction

Obstructive Sleep Apnea (OSA) is a chronic disorder marked by recurrent episodes of partial or complete upper airway obstruction during sleep. These repeated obstructions lead to sleep fragmentation and deprivation, which can disrupt normal brain function^1–3. Neuroimaging studies have shown that patients with OSA often exhibit white matter abnormalities in brain regions involved in autonomic regulation, emotional processing, cognitive control, and respiratory function⁴. Such structural changes are associated with common clinical symptoms of OSA, including excessive daytime sleepiness, depression, anxiety, and cognitive impairment³.

Diffusion tensor imaging (DTI) is an advanced MRI technique that characterizes the microstructural integrity of white matter by measuring the direction and magnitude of water molecule diffusion in neural tissue^5,6. DTI provides four key metrics: fractional anisotropy (FA), mean diffusivity (MD), axial diffusivity (AD), and radial diffusivity (RD). These parameters enable quantitative analysis of white matter integrity and can indicate axonal damage, loss, or demyelination by reflecting changes in axons and myelin within fiber tracts.

The Tract-Based Spatial Statistics (TBSS) is widely used for DTI analysis due to its low subjectivity and high accuracy⁷. Xiong et al.⁸ used TBSS to examine white matter changes in OSA patients undergoing CPAP, with or without residual sleep disturbances, and found that increased RD may indicate myelin damage. Similarly, Zhang et al.⁹ applied TBSS to compare OSA patients with good sleepers, revealing higher temporal diffusion heterogeneity α and anomalous diffusion coefficient D_m in the OSA group. While TBSS-based studies have identified white matter abnormalities in OSA, they often fail to clarify the relationship between structural changes and clinical symptoms, and findings remain inconsistent.

OSA-related brain damage often affects multiple regions, including impaired white matter integrity¹⁰, and may contribute to daytime sleepiness, depression, anxiety, and cognitive dysfunction¹¹. For effective investigation, an analytical method is needed that not only detects white matter abnormalities but also supports symptom-pathology correlation. Automated Fiber Quantification (AFQ) meets these requirements by enabling detection of tract-specific damage and direct associations between white matter changes and clinical symptoms (e.g., cognitive decline, anxiety, depression)¹². Therefore, by comparing the analysis method of TBSS and AFQ, this study can not only explore a more suitable data analysis method for the white matter changes of OSA patients, but also explore the local structural changes of white matter fibers in OSA patients, and explore the relationship between these local white matter structural changes and the clinical symptoms of the patients, and provide white matter microstructure-level insights into the neuropathological mechanism of clinical symptoms in OSA patients.

Materials and methods

Participants

This study was conducted in accordance with the Declaration of Helsinki and was approved by the Ethics Committee of the Affiliated Zhongshan Hospital of Dalian University (Approval No. 2020018). Written informed consent was obtained from all participants included in the study. All research procedures involving human participants were carried out in compliance with applicable Chinese regulations, including the Measures for Ethical Review of Life Sciences and Medical Research Involving Human Subjects.

Thirty-three patients with moderate-severe OSA (29 males, 4 females; mean age: 40.15 ± 8.52 years) were recruited from the Department of Otolaryngology at Zhongshan Hospital, Dalian University. OSA diagnosis was confirmed via overnight polysomnography (PSG, > 8 h) with an Apnea-Hypopnea Index (AHI) > 15 events/hour. Common symptoms included snoring, frequent awakenings, dry mouth, nocturia, daytime sleepiness, poor concentration, and memory decline. None had received prior treatment. During the same period, 28 healthy volunteers (21 males, 7 females; mean age: 37.86 ± 7.75 years) with AHI < 5 events/hour were enrolled as the No Sleep Disorder (NSD) control group.

Inclusion criteria for the OSA group were: (1) diagnosis based on the Obstructive Sleep Apnea Hypopnea Syndrome Diagnosis Guide (2011, modified) by the Sleep Apnea Group of the Chinese Respiratory Society (2012)¹³, with AHI > 15 events/hour confirmed by overnight PSG; recent diagnosis of moderate-to-severe OSA without prior treatment; (2) At least 9 years of education, right-handed, and body weight < 125 kg; (3) no history of mental or neurological disorders, or psychiatric medication use.

Exclusion criteria were: (1) diagnosis of sleep disorders other than OSA; (2) diagnosis of hypertension (blood pressure > 140/90 mmHg), diabetes, severe cardiovascular disease, or Alzheimer’s disease; (3) abnormal brain structural changes (e.g., severe white matter lesions, cerebrovascular lesions, cerebral atrophy, trauma, infarction, hemorrhage, or tumors); (4) contraindications for MRI.

Polysomnographic assessment

The polysomnography monitor (Philips Alice 6) was used to monitor the sleep of the subjects for the whole night. The whole process was performed in the sleep monitor room of the E.N.T Department of Affiliated Zhongshan Hospital of Dalian University. Sleep monitoring was performed prior to the head MRI examination. On the day of the monitoring, the patients were not allowed any stimulatory or hypnotic drugs.

Neuropsychological scales

All neuropsychological assessments (See supplementary file) were conducted prior to MRI acquisition. Participants completed the following standardized instruments in a fixed sequence:

The Epworth Sleepiness Scale (ESS) measures daytime sleepiness in eight situations, with total scores ranging from 0 to 24. Scores of 6, 11, and 16 indicate increasing severity from mild to severe sleepiness¹⁴.

The Self-Rating Anxiety Scale (SAS) includes 20 items rated on a 4-point Likert scale. After conversion (raw score × 1.25), scores below 50 are normal, 50–59 indicate mild anxiety, 60–69 moderate anxiety, and ≥ 70 severe anxiety¹⁵.

The Self-Rating Depression Scale (SDS) also includes 20 items, with scores divided by 80 to generate an index. Scores < 0.5 are normal, 0.5–0.59 indicate mild depression, 0.6–0.69 moderate depression, and ≥ 0.7 severe depression¹⁶.

MRI data collection

MRI imaging for all subjects was performed using a Siemens 3.0T Scanner with 12-channel phased array head coils (MagnetomVerio). The sponge mat was used to fix the head of the subjects to minimize head motion. For DTI data, the EPI sequence was triggered for one time (TR = 6000 ms; TE = 95 ms; flip angle = 90°; band width = 1502 Hz/pixel; matrix size = 128 × 128; FOV = 230 × 230 mm; slice thickness = 3.0 mm). Thirty diffusion gradient directions were set with b-values = 0, 1000 s/mm². The high-resolution sagittal view 3D-MPRAGE sequence was used to obtain T1-weighted thin layer image (TR/TE/flip angle=1900ms/2.79ms/9°; slice = 176; slice thickness = 1 mm; matrix size = 384 × 384; FOV = 259 mm×259 mm), so as to eliminate possible structural lesion and provide anatomical information for AC-PC positioning.

Data processing

DTI data pre-processing

The Oxford brain function MRI software FSL (http://www.fmrib.ox.ac.uk/fsl/) was then used to process the images¹⁷. Firstly, Eddy current correction and gradient correction are carried out on the original image to eliminate the deformation caused by head movement and eddy current during the scanning process, and then the original gradient direction was adjusted according to the changes due to eddy current correction. Secondly, the BET tool of FSL was used to extract the brain and removed non-brain structure for more accurate spatial matching. The range of analysis (fiber tracking) was limited to reduce the total calculation. Finally, the DTIFIT instructions were used on a diffusion tensor model to re-establish diffusion tensors, and each diffusion tensor parameter was calculated (FA, MD, RD, AD).

TBSS analysis method

(1) The linear or non-linear registration method was used to register the pre-processed individual FA maps to standard space. (2) Based on all FA registered to the standard space, the maps indicating average FA and the white matter skeleton were established. (3) The threshold value used for the average FA skeleton was 0.2. (4) The non-FA diffusion parameters (MD, AD, RD) skeleton was calculated. The FA maps for all subjects were registered to MD, AD, RD maps in the non-linear manner to generate the registered parametric map for projection onto the average FA fiber tract skeleton.

AFQ analysis method

The procedures for AFQ to identify and quantify fiber tracts are summarized below¹²:

1. A deterministic streamlines tracking algorithm (STT) was used to track the fiber tracts of whole brain of each subject^18,19. When the track record showed FA value < 0.2 or angle > 30°, the tracking was stopped.

2. Whole brain fiber tracts were divided using Waypoint ROI program²⁰. A fiber tract was defined by passage through two Waypoint ROIs at the same time. Wakana et al.²⁰, based on the descriptive definition from an anatomical perspective²¹, located each Waypoint in MNI standard space with group average DTI data. These ROI were transferred from MNI standard space to individual space through non-linear transfer.

3. The iteration frequency was used to eliminate the fiber tract technical error caused by data noise, complexity in fiber tract direction and area, and blurred stop criteria. Number of Iterations: The cleanup process is iterated until no more anomalous fibers are identified. In each iteration, the system checks all fibers in the current fiber bundle and determines if they are abnormal fibers based on statistical criteria. If abnormal fibers are present, remove them and move on to the next iteration; Otherwise, the iteration stops. Threshold for Retention: The identification of abnormal fibers is based on the Mahalanobis distance, calculated as follows:

In the formula: x represents the three-dimensional coordinates of a fiber at a specific node; µ denotes the average coordinates of that node (the core of the fiber bundle); S is the covariance matrix of that node. The Mahalanobis distance can be interpreted as a z-score in a multivariate Gaussian distribution, used to determine whether a point belongs to the distribution. If a fiber’s Mahalanobis distance exceeds 5 at any node, it is considered an outlier and should be removed. Additionally, fiber length serves as a screening criterion: fibers exceeding the mean by more than 4 standard deviations will also be removed.

Unit of frequency: The “frequency” here is not a unit of time, but rather an anomaly detection and removal operation performed in each iteration. In each iteration, the system recalculates the core position of the fiber bundle and the covariance matrix, reassessing the probability of attribution of all remaining fibers. This process continues until all fibers meet the retention conditions.

4) The Johns Hopkins University White-Matter Tractography Atlas we used mapped the voxel-level skeleton effects of TBSS onto fiber bundle labels. The 100 nodes with the same length were determined on each of the fiber tract, and according to the average location of each node, the core position of the fiber tract was determined. The characteristic values of diffusion tensor of each node was calculated (FA, MD, AD, RD) by taking the weighted average value.

Statistical analysis

SPSS 20.0 was used for statistical analysis. Between-group differences with respect to demographic characteristics, age, educational background, and body mass index were assessed using the independent sample t test. Chi-squared test was used to compare sex distribution in the two groups. Between-group differences in the TBSS data were assessed using nonparametric permutation test based on FSL. The threshold of statistical significance was set as p < 0.05. The MRI imaging indicators and neuropsychological indicators were regress out with covariates including gender. Age was not included as a covariate in the GLM design for TBSS analysis, as there was no significant age difference between groups (p = 0.279). Groups were well-matched on this demographic variable, and recent methodological studies suggest that including covariates without significant group differences can reduce statistical power without providing meaningful adjustment²². The Threshold Free Clustering Enhancement (TFCE) was used to compare and correct the statistical results for multiple comparisons.

For between-group analysis, AFQ method was used to gain the diffusion tensor indicators of each node of the 20 fiber tracts, and 500 permutations of random permutation tests were performed²³. Finally, the TFCE was used to compare and correct for multiple comparisons. The p values < 0.05 were considered indicative of statistical significance.

Results

Demographic and clinical data

Table 1 summarizes the demographic characteristics, sleep parameters, and neuropsychological outcomes for both the NSD and OSA groups. There were no statistically significant differences between the groups in terms of age, gender, body mass index (BMI), educational attainment, Self-Rating Anxiety Scale (SAS), or Self-Rating Depression Scale (SDS) scores (p > 0.05). However, individuals with OSA exhibited significantly elevated values for BMI, Epworth Sleepiness Scale (ESS), apnea-hypopnea index (AHI), and micro-arousal index relative to their NSD counterparts (p < 0.05). In contrast, participants in the NSD group had higher minimum oxygen saturation (Mini SpO₂) levels (p < 0.05).

Table 1.

Results of the three groups in demographic data, sleep monitoring and neuropsychological scale tests.

Variable	OSA (N = 33)	NSD (N = 28)	t/χ2	p
Age (years old)	40.15 ± 8.52	37.86 ± 7.75	-1.401	0.279
Sex (male/female)	29/4	21/7	-	-
BMI (kg/m2)	27.72 ± 3.31	24.56 ± 3.09	-5.314	0.01*
Education (years)	14.88 ± 2.13	15.54 ± 2.06	1.465	0.229
SAS	40.28 ± 8.10	39.35 ± 5.57	-0.512	0.611
SDS	0.38 ± 0.10	0.37 ± 0.08	-0.308	0.759
ESS	10.64 ± 6.21	7.71 ± 3.79	-2.168	0.034*
AHI (events/h)	2.02 ± 1.19	46.36 ± 22.57	-10.371	< 0.001*
Mini SpO2 (%)	0.79 ± 0.10	0.92 ± 0.03	7.181	< 0.001*
Mean SpO2 (%)	0.94 ± 0.02	0.97 ± 0.01	5.766	< 0.001*
MI (events/h)	35.77 ± 18.16	20.05 ± 15.21	-3.626	< 0.001*

Abbreviations: BMI, body mass index; SAS, self-rating for anxiety; SDS, self-rating for depression; ESS, Epworth Sleepiness Scale; AHI, apnea hypopnea index; MI, micro-arousal index; The * represents p < 0.05.

TBSS: between-group difference in diffusion tensor indicators

Nonparametric permutation testing based on FSL revealed that the OSA group exhibited reduced MD in the forceps minor, bilateral anterior thalamic radiations, and the left superior longitudinal fasciculus relative to the NSD group (Fig. 1A). Additionally, AD was significantly lower in the OSA group within the forceps minor, left corticospinal tract, left anterior thalamic radiation, left superior longitudinal fasciculus, and left inferior longitudinal fasciculus compared to the NSD group (Fig. 1B). Nevertheless, none of these differences remained statistically significant following threshold-free cluster enhancement (TFCE) correction. Furthermore, analysis of FA and RD values showed no significant alterations in any fiber tracts between groups, with or without statistical correction (Fig. 1C and D).

Fig. 1.

Tract-based spatial statistics (TBSS) of MD, AD, FA and RD between OSA and NSD groups. A-B: The MD and AD skeleton is displayed in green, which represent the center of the white matter tracts. The red areas represent the fiber tracts that show decreased value of MD and AD compared with the control group (p < 0.05, without TFCE corrected). C-D: The FA skeleton is displayed in green and RD skeleton is displayed in red, which represent the center of the white matter tracts. There were no abnormal brain regions on FA and RD fiber skeleton with or without TFCE correction. 1: forceps minor (-13,55,-11); 2: left anterior thalamic radiations (-17,-19,7); 3: right anterior thalamic radiations (21,49,23); 4: left superior longitudinal fasciculus (-45,-39,41); 5: forceps minor (-13,50,26); 6: left corticospinal tract (-9,-18,-18); 7: left anterior thalamic radiations (-5,-17,-2); 8: left inferior longitudinal fasciculus (-52,-17,-2); 9: left inferior longitudinal fasciculus (-13,-71,41); 10: left superior longitudinal fasciculus (-46,-50,41).

Difference in dispersion tensor indicators at AFQ fasciculus segment

Fractional anisotropy

Following TFCE-corrected nonparametric permutation testing using FSL, the OSA group demonstrated significantly lower FA values at nodes 77–78 within the callosal forceps major (Fmajor), when compared to the NSD group (Fig. 2A; Table 2).

Fig. 2.

Relative to the NSD group, the OSA group exhibited reduced FA values at nodes 77–78 in the callosal forceps major (Fmajor; A), elevated MD values at nodes 23–24 in the right thalamic radiation (TR.R; B), and increased AD values at nodes 54–57 in the right inferior fronto-occipital fasciculus (IFOF.R; C) as well as nodes 66–68 in the right superior longitudinal fasciculus (SLF.R; D). In all four panels, green lines denote the NSD group and red lines indicate the OSA group. Solid lines represent group means, while dotted lines reflect the standard deviation. The X-axis spans 1 to 100 fiber nodes, and the Y-axis corresponds to FA values. The X-axis represents 100 equidistant nodes along each fiber tract, and the Y-axis represents the corresponding diffusion metric values. The color bar represents TFCE-corrected P-values, ranging from blue to red corresponds to a P-value range of 0 to 1.

Table 2.

Difference of FA at Fmajor, MD at TR.R, and AD at the IFOF.R and SLF.R between the NSD and OSA groups.

	Node	NSD	OSA	t	p
Fmajor	77	0.579 ± 0.517	0.530 ± 0.673	3.163*	0.042
	78	0.581 ± 0.521	0.530 ± 0.715	3.116*	0.044
TR.R	23	0.688 ± 0.192	0.706 ± 0.305	2.775*	0.048
	24	0.689 ± 0.192	0.708 ± 0.294	2.813*	0.042
IFOF.R	54	1.151 ± 0.606	1.168 ± 0.219	3.277*	0.032
	55	1.153 ± 0.597	1.173 ± 0.218	3.661*	0.016
	56	1.154 ± 0.611	1.173 ± 0.218	3.630*	0.016
	57	1.153 ± 0.627	1.170 ± 0.219	3.275*	0.032
SLF.R	66	1.091 ± 0.772	1.138 ± 0.645	2.566*	0.046
	67	1.091 ± 0.770	1.137 ± 0.640	2.580*	0.042
	68	1.090 ± 0.767	1.137 ± 0.636	2.567*	0.046

Abbreviations: Callosum Forceps Major, Fmajor; Right Thalamic Radiation, TR.R; Right inferior fronto-occipital fasciculus, IFOF.R; Right Superior Longitudinal Fasciculus, SLF.R. * indicates p < 0.05.

Mean diffusivity

After applying TFCE correction in the nonparametric permutation test conducted via FSL, the OSA group exhibited elevated MD values at nodes 23–24 within the right thalamic radiation (TR.R), relative to the NSD group (Fig. 2B; Table 2).

Axial diffusivity

Based on the FSL nonparametric permutation analysis, the OSA group exhibited higher AD values than the NSD group at nodes 54–57 of the right inferior fronto-occipital fasciculus (IFOF.R) and nodes 66–68 of the right superior longitudinal fasciculus (SLF.R) (Fig. 2C–D; Table 2).

Radial diffusivity

The TFCE-corrected nonparametric permutation analysis revealed no fiber nodes with statistically significant differences on RD values between the OSA and NSD groups (Table 3).

Table 3.

The difference of RD value in each WM fiber between the NSD and OSA groups.

	NSD (n = 28)	OSA (n = 33)	t	p
ATR_L	0.543 ± 0.238	0.548 ± 0.275	0.771	0.444
ATR_R	0.532 ± 0.256	0.529 ± 0.271	0.515	0.609
CST_L	0.429 ± 0.242	0.429 ± 0.223	0.040	0.968
CST_R	0.437 ± 0.216	0.442 ± 0.228	0.799	0.428
CCing_L	0.556 ± 0.456	0.556 ± 0.481	0.050	0.960
CCing_R	0.552 ± 0.336	0.512 ± 0.140	1.477	0.145
CHippo_L	0.253 ± 0.322	0.253 ± 0.321	0.011	0.991
CHippo_R	0.436 ± 0.308	0.308 ± 0.323	1.583	0.119
Fmajor	0.521 ± 0.079	0.519 ± 0.084	0.109	0.913
Fminor	0.500 ± 0.039	0.492 ± 0.038	0.841	0.404
IFOF_L	0.584 ± 0.027	0.579 ± 0.027	0.592	0.557
IFOF_R	0.582 ± 0.023	0.564 ± 0.110	0.820	0.416
ILF_L	0.633 ± 0.035	0.625 ± 0.039	0.806	0.423
ILF_R	0.638 ± 0.036	0.635 ± 0.042	0.279	0.781
SLF_L	0.559 ± 0.038	0.559 ± 0.030	0.062	0.951
SLF_R	0.532 ± 0.031	0.525 ± 0.390	0.787	0.434
UF_L	0.649 ± 0.034	0.637 ± 0.054	1.113	0.270
UF_R	0.618 ± 0.038	0.607 ± 0.046	1.010	0.316
AF_L	0.524 ± 0.033	0.521 ± 0.037	0.245	0.807
AF_R	0.440 ± 0.185	0.463 ± 0.153	0.529	0.599

Abbreviations: ATR, anterior thalamic; CST, corticospinal tract; CCing, cingulum cingulate; CHippo, cingulum hippocampus; Fmajor, Callosum Forceps Major; Fminor, Callosum Forceps Major; IFOF, inferior fronto-occipital fasciculus; ILF, inferior longitudinal fasciculus; SLF, superior longitudinal fasciculus; UF, uncinated fasciculus; AF, arcuate fasciculus. L, left; R, right. p < 0.05 indicates statistically significant difference.

Correlation between white matter changes and clinical indicators in OSA

White matter fiber bundles that exhibited significant differences in FA, MD, AD, or RD values between the NSD and OSA groups were designated as regions of interest for subsequent correlation analyses. Pearson correlations were computed between diffusion metrics (FA, MD, AD, and RD) of these tracts and clinical measures in patients with OSA. Notably, within the OSA group, FA values in the callosal forceps major (Fmajor) were positively associated with SAS scores at node 77 (r = 0.422, p = 0.014) and node 78 (r = 0.447, p = 0.009) (Fig. 3A). In addition, AD values in the right inferior fronto-occipital fasciculus (IFOF.R) demonstrated a significant negative correlation with SAS scores at node 55 (r = -0.346, p = 0.049), node 56 (r = -0.406, p = 0.019), and node 57 (r = -0.388, p = 0.026) (Fig. 3B). Furthermore, AD values in the same tract were inversely correlated with SDS scores at node 54 (r = -0.521, p = 0.002), node 55 (r = -0.516, p = 0.002), and node 56 (r = -0.426, p = 0.013) (Fig. 3B).

Fig. 3.

A shows the correlation between FA at Fmajor and anxiety score. B shows the correlation between AD values at IFOF.R and anxiety/depression scores. The blue plot means that anxiety/depression scores are distributed across participants. The yellow plot means that diffusion indices of each fiber node are distributed across participants.

Discussion

This study compared AFQ and TBSS methods, finding that AFQ more effectively detected white matter fiber changes in OSA patients due to its higher spatial resolution. In addition to identifying changes linked to hypoxia and sleep fragmentation, we also uncovered associations between white matter changes and anxiety and depression. Given the prevalence of anxiety and depression in OSA patients, which may exacerbate brain damage, future research should focus not only on the effects of traditional physiological factors like hypoxia and sleep fragmentation but also on the impact of mental health conditions such as anxiety and depression on brain changes.

Comparison of AFQ and TBSS methods in DTI data of OSA patients

Although previous studies using TBSS have identified white matter changes in OSA patients^8,9, there is heterogeneity in their results, and the relationship between white matter changes and clinical symptoms remains unclear. In contrast to TBSS, which relies on spatial averages that may mask focal changes, AFQ allows node analysis at 100 equidistant points on each bundle, enabling precise localization of microstructural abnormalities. This capability enhances our ability to explore the relationship between brain changes and clinical symptoms.

This study is the first to apply Automated Fiber Quantification (AFQ), a high-resolution tract-specific DTI analysis method, to examine white matter alterations in OSA patients²³, alongside the conventional Tract-Based Spatial Statistics (TBSS) approach. Before threshold-free cluster enhancement (TFCE) correction, the OSA group showed reduced MD and AD in several tracts, including the forceps minor, anterior thalamic radiations, and the superior and inferior longitudinal fasciculi. However, these differences did not survive TFCE correction, possibly due to the limited sample size or the subtle nature of the white matter changes.

After applying the AFQ method, nodes 77–78 in the callosum forceps major showed decreased functional anisotropy, nodes 23–24 in the right thalamic radiation showed increased MD, and nodes 54–57 in the right inferior fronto-occipital fasciculus and nodes 66–68 in the right superior longitudinal fasciculus showed increased AD, compared to the NSD group (p < 0.05, corrected by threshold-free cluster enhancement). These findings support the superior suitability of AFQ over TBSS for detecting subtle white matter changes in OSA patients.

The White Matter micro-structure changes in OSA

The MD, reflecting the overall magnitude of water diffusion²⁴, is influenced by both AD and RD and is sensitive to cellular density, edema, and necrosis²⁵. Its temporal dynamics enable differentiation of ischemic stages: MD decreases in the acute phase, normalizes in the subacute phase, and increases in the chronic phase²⁶. In our cohort, primarily consisting of individuals with moderate-to-severe OSA and longer disease duration, we observed elevated MD in the right thalamic radiation (TR.R), consistent with chronic white matter pathology²⁷.

The AD reflects axonal integrity and is influenced by cellular swelling and extracellular space reduction²⁸. Recurrent hypoxia and reoxygenation cycles in OSA disrupt ionic homeostasis, leading to cytotoxic edema and restricted diffusion. In this study, we identified increased AD in the right inferior fronto-occipital fasciculus (IFOF.R) and superior longitudinal fasciculus (SLF.R), consistent with previous studies^29,30.

The SLF is crucial for visuospatial processing, attention, eye movement control, and integration of parietal and frontal cortical activity. Microstructural changes in the SLF in OSA patients may impair attention, memory, visuospatial function, and executive processing³¹. Additionally, previous studies have shown that insomnia can alter SLF diffusion indicators and thalamic volume, leading to decreased attention and slower processing speed³². In this study, we also observed increased AD in the superior longitudinal fasciculus (SLF.R), suggesting that the change in SLF may be driven by sleep quality decline, consistent with previous findings linking insomnia to changes in SLF diffusion indices³².

The FA is widely regarded as a marker of white matter fiber integrity³³. In OSA, small-caliber pial arteries in deep white matter (~ 100–400 μm) are poorly collateralized, making them vulnerable to ischemia and hypoxia, which may reduce FA in associated tracts. Previous studies have identified abnormalities in the corpus callosum of OSA patients³⁴, including reduced FA in the anterior part of the corpus callosum³⁵, cingulate gyrus, and corticospinal tract, decreased gray matter volume in the corpus callosum region³⁶, and lower entropy values in the posterior corpus callosum³⁷. These findings align with the significant reduction in FA values observed in the large claw portion of the corpus callosum in this study.

Relationship between the White Matter Changes and Anxiety/Depression in OSA

In this study, FA in the forceps major was positively associated with anxiety severity, whereas AD in the right inferior fronto-occipital fasciculus was negatively associated with anxiety. Both tracts involve occipital regions implicated in visual processing. Prior work suggests that heightened occipital cortical structure and activity are associated with increased anxiety^38–40, and that reductions in occipital engagement accompany symptom improvement⁴¹. Together, these findings raise the possibility that disruption of occipital white-matter pathways may attenuate anxiety-related neural processing in OSA, although the underlying mechanisms remain speculative.

The right prefrontal lobe primarily processes negative emotions (e.g., fear, disgust, anxiety), while the left prefrontal lobe regulates positive emotions⁴². The IFOF.R, as a long association fiber bundle, plays a role in interbrain information integration. Some of its fibers are indirectly connected to the temporal lobe structures, such as the hippocampus and amygdala, through adjacent pathways like the hook bundle. Damage to the white matter microstructure of the right frontal-occipital fasciculus (IFOF.R) may disrupt the connection between the right prefrontal cortex and the amygdala, impairing the transmission of hyperactive negative emotional signals, thus reducing anxiety and depression. In this study, AD values at IFOF.R showed a significant negative correlation with SAS and SDS scores. We speculate that IFOF.R damage blocks the transmission of excessive negative emotional signals (such as disrupted connections between the amygdala and prefrontal cortex), which may lead to reduced anxiety and depression.

Limitations and future works

Some limitations of this study should be acknowledged. First, the small sample size and the limited number of female participants may introduce bias. Larger studies are needed to provide more generalizable findings on microstructural changes in the white matter fiber tracts of OSA patients. Secondly, a subgroup analysis based on neurological scales (e.g., Self-Rating Anxiety Scale, Self-Rating Depression Scale) in a larger cohort, combined with clinically relevant cognitive assessments, could offer better neuroimaging markers for clinical application. Finally, smoking status/tobacco exposure is a significant risk factor for OSA^43–45. To mitigate confounding effects and facilitate robust statistical modeling, future studies must systematically collect comprehensive tobacco exposure data—including current, past, never status, intensity (e.g., pack-years), duration, and product types—and incorporate these variables as key covariates in analyses.

Additionally, in the field of image processing technology, we acknowledge that our preprocessing pipeline omitted advanced denoising and corrections for Gibbs ringing and gradient inhomogeneities, which are now considered best practices^46–48. Nevertheless, our standardized preprocessing using FSL’s eddy correction constitutes a well-validated approach employed successfully in numerous published DTI studies. Recent methodological investigations indicate that while supplementary preprocessing steps can enhance accuracy, minimally processed data remain valid for clinical research applications^49,50. Future studies should incorporate these advanced techniques to further improve data quality.

The AFQ pipeline utilizes deterministic tractography, which can be limited in regions exhibiting complex fiber configurations. Nonetheless, extensive validation studies have established AFQ’s reliability for detecting segmental white matter abnormalities in clinical cohorts^51,52. This method has been successfully applied across diverse neurological disorders. Future research should investigate whether probabilistic tractography offers enhanced sensitivity for identifying white matter alterations in obstructive sleep apnea (OSA).

Conclusion

The TBSS method failed to detect white matter microstructure changes in OSA patients, possibly due to the limited sample size ‌leading to‌ biased results, which does not negate the ‌application‌ of TBSS technology in detecting brain white matter changes in OSA patients. However, in the same sample, AFQ technology, with its higher resolution, ‌was more sensitive than‌ TBSS in detecting white matter microstructure changes in OSA patients. This technique opens new avenues for exploring white matter changes in OSA patients. Significant alterations were observed in the corpus callosum, right inferior fronto-occipital bundle, right superior longitudinal fasciculus, and right thalamic radiation in OSA patients. The levels of anxiety and depression in OSA patients were linked to microstructural changes in the corpus callosum and right inferior fronto-occipital bundle, offering white matter-level insights into the neuropathological mechanisms of anxiety and depression in these patients. Although OSA could increase patients’ anxiety and depression levels, the loss of white matter nerve fiber integrity could alleviate patients’ anxiety and depression levels. Howere, brain white matter damage in OSA patients appears to reduce anxiety and depression, this should not be seen as a positive outcome. At this stage, the damage is relatively mild, and the affected areas may inadvertently reduce anxiety and depression levels. However, as white matter damage progresses, it may lead to more severe neurodegenerative conditions, such as cognitive impairment.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Concept and design: Danyang, Li and Jianlin, Wu; data collection and analysis: Danyang, Li , Xiangbo, Yan , and Wei, Zhao ; drafting of the article: Danyang, Li , Xiangbo, Yan , and Ningning, An; critical revision of the article for important intellectual content: Jianlin, Wu and Qing, Zhang; study supervision: Jianlin, Wu and Qing, Zhang. All the authors approved the final article.

Funding

This study received no special funding support.

Data availability

Our data was published on the public database OSF (DOI: 10.17605/OSF.IO/5Q4B8) and could be found here：https://osf.io/5q4b8/files/osfstorage? view_only=59a89bbc1f7d4322a1a96ef1c907acdb.

Competing interests

The authors declare no competing interests.

Ethical approval

This study has been approved by the Ethics Committee of Affiliated Zhongshan Hospital of Dalian University (the ethics approval number: 2020018), and all subjects signed the informed consent before the experiment.

Informed consent

Informed consent was obtained from all individual participants included in the study.

Open Practices

The design and analysis plan for the study were pre-registered at https://osf.io/5q4b8/?view_only=59a89bbc1f7d4322a1a96ef1c907acdb. The results meet our hypothesis pre-registered.