Cortical functional and structural alterations in rhegmatogenous retinal detachment: a multimodal analysis using surface-based independent component analysis, morphometry, and machine learning

Introduction

Rhegmatogenous retinal detachment (RRD) occurs when fluid vitreous leaks through retinal tears, separating the retina’s neural and pigment epithelial layers (1). The most frequent cause of RRD is age-related posterior vitreous detachment; however, other etiological factors, such as high myopia, post-operative changes following cataract surgery, and ocular trauma, may also contribute to its development (2,3). In its early stages, RRD typically presents with symptoms, including floaters, flashes, and a loss of visual fields. As RRD progresses, its extension to the macular region can drastically reduce vision and potentially lead to blindness without prompt and effective treatment. Worldwide, RRD affects about 12.17 per 100,000 individuals, with the highest rates reported in Europe. Estimates predict that these numbers will double in the upcoming years (4).

Despite its high occurrence and serious impact on vision, full recovery post-surgery is rare due to likely irreversible damage to photoreceptors and visual fields (5-7). The retina, which is part of the central nervous system (8), has a nerve fiber thickness that is correlated with changes in the visual pathways and the brain’s visual cortex regions (9). Advances in neuroimaging have shown abnormal activity in the visual pathways and related gray matter areas in RRD patients (10-12). Previous studies on RRD have largely focused on volumetric or voxel-based alterations in subcortical and global gray matter structures, while the cortical surface, particularly its network-level functional and structural organization, has received comparatively less attention.

Independent component analysis (ICA) is a data-driven technique that uses blind source separation to process datasets. The method aims to decompose data into statistically independent spatial and temporal components, allowing for the precise extraction of brain activity signals and effective noise removal (13-15). ICA has been widely applied in neuroscience to reveal large-scale functional organization in both healthy and abnormal brains with considerable success (16-18). However, most of these studies have relied on voxel-based analysis. Research indicates that brain function is organized along cortical surfaces (19), and the cortical topological features make traditional voxel-based magnetic resonance imaging (MRI) inadequate for representing its layered structure. Conversely, surface-based cortical analysis provides a more accurate representation of the cortical structure, enhancing segmentation, the signal-to-noise ratio, and algorithm reproducibility (20,21). Thus, surface-based ICA provides a more reliable spatial framework for resting-state functional networks, allowing independent components (ICs) to better align with the brain’s functional regions.

Surface-based morphometry (SBM) analysis is an imaging technique that analyzes brain structural MRI data by quantifying cortical features like thickness, surface area, volume, and curvature (22,23). Cortical thickness (CT) is a key measure of neuronal density and cortical integrity (24,25), and is more sensitive and direct than gray matter volume in revealing pathophysiological mechanisms (26). Previous voxel-based morphometry studies have shown reduced gray-white matter volume in the visual regions of RRD patients (27), but these volumetric approaches did not assess surface-based cortical features and relied on unidimensional analyses, which may lack spatial specificity and reliability. We sought to combine surface-based ICA with SBM to provide a more comprehensive view of the cortical function and structure of RRD patients.

To further validate the consistency and reliability of our results, we used support vector machine (SVM), a powerful machine learning algorithm that effectively transforms high-dimensional neuroimaging data into clinically valuable decision-making tools (28,29). SVM creates an optimal separating hyperplane to maximize the margin between classes and employs the Gaussian radial basis function (RBF) (30) kernel to map nonlinearly separable data into a higher-dimensional space, making it linearly separable (31,32). However, due to limited interpretability of the predictions, the use of SVM alone faces challenges in clinical applications. To address this, we integrated the SHapley Additive exPlanations (SHAP) method (33) to evaluate the importance of each feature in our final model, improving its interpretability.

We hypothesized that sudden visual loss in RRD patients may be associated with changes in functional connectivity (FC), leading to subtle cortical structural adaptations rather than rapid or pronounced structural remodeling. To test this hypothesis, we used surface-based ICA and SBM techniques to segment the cortical network of RRD patients, identifying five resting-state cortical functional networks. We then used the SBM method to calculate CT across the entire brain, identifying regions with statistically significant differences between the RRD patients and healthy controls (HCs). To conclude, we assessed the discriminative power of neuroimaging-derived features using SVM models and employed SHAP to identify the cortical regions that contributed most to the classification outcomes. This integrative framework offers solid neuroimaging-based confirmation of cortical alterations in RRD patients and extends our understanding of the disorder’s neuropathological underpinnings, while providing fresh perspectives for clinical assessment and therapeutic planning. We present this article in accordance with the STROBE reporting checklist (available at https://qims.amegroups.com/article/view/10.21037/qims-2025-628/rc).

Methods

Participants

This study was approved by the Medical Ethics Committee of The First Affiliated Hospital of Nanchang University (approval No. IIT [2024] Ethics No. 790), and all participants provided written informed consent. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments.

The study included 42 RRD patients and 45 HCs. RRD patients were included in the study if they met the following inclusion criteria: (I) had idiopathic RRD with one or more retinal tears; (II) had RRD affecting one or more quadrants; and (III) had no other ocular diseases in either eye. The majority of the RRD patients presented with sudden-onset visual decline and had no identifiable predisposing factors, and were thus classified as idiopathic RRD patients. Patients were excluded from the study if they had a history of recurrent RRD, RRD secondary to high myopia, ocular trauma, diabetes, previous fundus surgery, cardiovascular or neuropsychiatric disorders, or cerebrovascular events. HCs were recruited from the local population in Nanchang and matched to the RRD patients in terms of age, gender, and educational level. HCs were included in the study if they met the following inclusion criteria: (I) had no ocular or major systemic diseases; (II) had a best-corrected visual acuity of 1.0 or better; and (III) had undergone MRI, optical coherence tomography (OCT), and ocular B-scan ultrasonography.

Functional MRI (fMRI) data acquisition

The fMRI data were acquired using a 3.0 Tesla MRI scanner (Siemens Trio Tim, Erlangen, Germany) equipped with an 8-channel phased-array head coil at The First Affiliated Hospital of Nanchang University. Prior to scanning, the participants were asked to lie in a supine position with their eyes closed and remain calm. A monitoring system was employed to confirm that the participants stayed awake and did not engage in active thought. To reduce ambient noise, earplugs were used, and foam cushions were positioned around the head to minimize motion artifacts. After image acquisition, participants’ psychological status was briefly evaluated to confirm procedural safety. A detailed overview of the experimental procedure is shown in Figure 1, and the scanning parameters are summarized in Table 1.

Figure 1 Schematic diagram of the analysis workflow of this study. (A) Perform group ICA and identify ICNs. (B) Intra-cortical network FC analysis. (C) Inter-cortical network FC analysis. (D) Surface-based morphometry analysis. (E) SVM analysis. (F) SHAP analysis for model explanation. AN, auditory network; CT, cortical thickness; DAN, dorsal attention network; DMN, default mode network; FC, functional connectivity; ICA, independent component analysis; ICN, independent component network; lh_lateraloccipital, left hemisphere lateral occipital cortex; rh_lateralorbitofrontal, right hemisphere lateral orbitofrontal cortex; rh_pericalcarine, right hemisphere pericalcarine cortex; SHAP, SHapley Additive exPlanations; SMN, sensorimotor network; SVM, support vector machine; VN, visual network.

Cortical functional data preprocessing

The resting-state fMRI data were preprocessed using fMRIPrep (34) (version 23.1.4) with surface-based outputs (-cifti-output 91k) aligned to the freesurfer left-right (fsLR) 32k space. The preprocessing pipeline followed a standard order: (A) slice timing correction was applied to adjust for the interleaved acquisition of fMRI volumes; (R) motion correction was conducted using MCFLIRT to realign all volumes to a reference volume; (W) spatial normalization involved the co-registration of functional images to the T1-weighted anatomical image, followed by nonlinear warping to the MNI152NLin2009cAsym space and projection onto the fsLR 32k standard surface; (S) spatial smoothing was performed post-fMRIPrep using Connectome Workbench, where a 6 mm full-width at half-maximum (FWHM) Gaussian kernel was applied along the mid-thickness surface mesh using wb_command-metric-smoothing. After smoothing, the left and right hemispheres were extracted from the Connectivity Informatics Technology Initiative (CIFTI) file using wb_command-cifti-separate, and then recombined to generate 64 k fsLR dense time series files (fsLR_den-64k_bold_smoothed.dtseries.nii), which served as the input for the ICA.

Processing of surface-based ICA

The preprocessed functional images from fMRIPrep (34) were first organized into a separate folder, and a list of CIFTI files (*.dtseries.nii) was created. Surface-based ICA was then performed using the FSL MELODIC toolbox (15). The functional data were mapped to the cortical surface and decomposed into ICs. At the group level, the number of ICA components was set to 30, a parameter choice widely adopted in resting-state fMRI studies that balances model stability and resolution while minimizing the risk of overfitting or underfitting (35). Among the extracted ICA components, five canonical functional networks were identified by combining a visual inspection with spatial pattern comparison against established templates as described previously (18,36,37). These included the visual network (VN), default mode network (DMN), auditory network (AN), dorsal attention network (DAN), and sensorimotor network (SMN). The resulting IC files were converted to CIFTI format and visualized using the wb_command tool to identify functionally relevant components. The spatial distribution and time series of each component were then back-projected to the individual level using inverse reconstruction (18,38,39), followed by statistical validation with a general linear model. The specific networks and component maps are shown in Figure 2.

Figure 2 Eleven resting-state cortical functional networks. AN, auditory network; DAN, dorsal attention network; DMN, default mode network; IC, independent component; SMN, sensorimotor network; VN, visual network.

SBM analysis processing

The T1 structural image data from all participants were processed using MATLAB 2017b (https://www.mathworks.com/products/matlab.html), along with the SPM12 (https://www.fil.ion.ucl.ac.uk/spm/software/spm12/) and CAT12 (https://neuro-jena.github.io/cat//) toolboxes. The processing pipeline included the following steps: automated segmentation of gray matter, white matter, and cerebrospinal fluid for each participant’s structural images. CT was then estimated using the projection-based thickness method (22), followed by surface reconstruction. This method effectively addresses partial volume effects, cortical blurring, and asymmetry without requiring explicit reconstruction of the sulci (22). Surface topology was subsequently corrected using spherical harmonic functions (40). The DARTEL algorithm (41) was used for spatial normalization, mapping the segmented tissue maps to the standard Montreal Neurological Institute (MNI) space (42). Additional processing was performed using spherical mapping (43). After quality checks, the CT measurements were calculated based on the reconstructed cortical surfaces. Finally, the measured surface data were resampled and smoothed with a 15-mm FWHM Gaussian kernel.

SVM and SHAP analysis

To assess whether cortical functional and structural indicators can distinguish RRD patients from HCs, we extracted these features from the dataset for classification analysis. Using MATLAB 2017b (https://www.mathworks.com/products/matlab.html), along with the libsvm-3.24 and svm_function toolboxes (31), we divided the dataset into training and testing sets to prepare for model construction. During model optimization, we chose the RBF kernel (44) and conducted hyperparameter tuning through grid search to identify the optimal hyperplane. To ensure accurate evaluation, we used leave-one-out cross-validation, where each sample served as an independent validation set. We then evaluated the classifier’s performance using several metrics, including accuracy, sensitivity, specificity, precision, and the area under the curve (AUC).

To enhance the interpretability of the SVM model’s predictions, we used the SHAP package in Python to assess the contribution of each feature to the model’s output. The SHAP algorithm, grounded in game theory, assigns an importance value to each feature, quantifying its contribution to the model’s prediction and offering interpretability for “black-box” machine learning models.

Correlation analysis

To explore the potential clinical relevance of cortical functional alterations, we performed correlation analyses between the regional FC values and preoperative clinical variables. Two indicators reflecting the severity of RRD were considered: the duration of retinal detachment (in days, based on clinical history), and the height of macular foveal detachment (measured via OCT).

The mean FC values were extracted from two VN regions, VN1 and VN2, which showed abnormal alterations in RRD patients. Since the clinical variables were not normally distributed, Spearman’s rank correlation coefficient was used to assess the relationships between the FC values and clinical measures.

All analyses were conducted using SPSS 27 (SPSS Inc., Chicago, IL, USA). A two-tailed P value <0.05 was considered statistically significant.

Statistical analysis

The statistical analyses were performed using SPSS version 27 (SPSS Inc.). To compare the clinical characteristics between the RRD and HC groups, independent two-sample t-tests were employed. For cortical functional measures, statistical maps were corrected using random field theory (RFT) with a voxel-level threshold of P<0.001 and cluster-level P<0.05. For cortical morphometric data, family-wise error (FWE) correction was applied (voxel P<0.01, cluster P<0.05). In all analyses, age, sex, and years of education were included as covariates to enhance the robustness and interpretability of the results.

Results

Demographic characteristics

No significant differences in gender and age were observed between the two groups of patients. Detailed information is provided in Table 2.

Analysis of FC within resting-state cortical functional networks

The RRD patients exhibited significantly higher FC values in the VN, DMN, AN, DAN, and SMN than the HCs. These changes in FC were primarily localized to the following brain regions: L_Visual_1, L_Visual_18, R_Default_29, R_Default_40, L_Auditory_10, L_Auditory_7, L_DorsalAttn_11, L_DorsalAttn_14, L_SMhand_10, L_SMhand_18, and L_SMmouth_1. These findings are presented in Figure 3 and Table 3 with RFT correction (voxel P<0.001, cluster P<0.05).

Figure 3 Distribution map of brain regions with significant FC differences in cortical resting-state networks in both groups. *, significant. AN, auditory network; DAN, dorsal attention network; DMN, default mode network; FC, functional connectivity; HCs, healthy controls; IC, independent component; L, left; R, right; RRD, rhegmatogenous retinal detachment; SMN, sensorimotor network; VN, visual network.

Analysis of functional network connectivity (FNC) between resting-state cortical functional networks

FNC refers to the temporal correlations between spatially independent resting-state networks, representing functional interactions between distinct networks. The RRD group exhibited increased FNC between the VN and SMN compared to the HC group. These findings are illustrated in Figure 4 (P<0.01).

Figure 4 Brain networks with significant FNC differences between the two groups of cortical resting-state networks and their corresponding matrix diagrams. AN, auditory network; DAN, dorsal attention network; DMN, default mode network; FNC, functional network connectivity; HCs, healthy controls; RRD, rhegmatogenous retinal detachment; SMN, sensorimotor network; VN, visual network.

CT results

The RRD group exhibited significant reductions in CT in the left hemisphere lateral occipital cortex (lh_lateraloccipital), right hemisphere pericalcarine cortex (rh_pericalcarine), and right hemisphere lateral orbitofrontal cortex (rh_lateralorbitofrontal) regions compared to the HC group. Detailed results are presented in Figure 5 and Table 4 with FWE correction applied (voxel P<0.01, cluster P<0.05).

Figure 5 Distribution map of brain regions with significant CT differences in both groups. The dotted circles indicate the locations of brain regions with significant cortical thickness differences (positive regions). *, significant. CT, cortical thickness; HCs, healthy controls; lh_lateraloccipital, left hemisphere lateral occipital cortex; rh_lateralorbitofrontal, right hemisphere lateral orbitofrontal cortex; rh_pericalcarine, right hemisphere pericalcarine cortex; RRD, rhegmatogenous retinal detachment.

SVM and SHAP results

Table 5 displays the evaluation results of the SVM model based on cortical functional and structural indicators, including the AUC, accuracy, sensitivity, specificity, and precision. The corresponding receiver operating characteristic curve is illustrated in Figure 6A. The results of the SHAP analysis for the SVM model are presented in Figure 6B-6D. Notably, based on the combined SVM and SHAP analysis, the cortical FC value of VN1 stands out. Specifically, the FC value of VN1 achieved high accuracy in the SVM model (accuracy =0.701), with its average absolute SHAP value being significantly higher than that of the other features.

Figure 6 SVM and SHAP analysis combining cortical functional and structural indicators. (A) Receiver operating characteristic curves of the SVM classifier. (B) SHAP summary plot. (C) SHAP value heatmap. (D) Comparison of average absolute SHAP values between RRD and HCs. AN, auditory network; CT, cortical thickness; DAN, dorsal attention network; DMN, default mode network; FC, functional connectivity; HCs, healthy controls; lh_lateraloccipital, left hemisphere lateral occipital cortex; rh_lateralorbitofrontal, right hemisphere lateral orbitofrontal cortex; rh_pericalcarine, right hemisphere pericalcarine cortex; RRD, rhegmatogenous retinal detachment; SHAP, SHapley Additive exPlanations; SMN, sensorimotor network; SVM, support vector machine; VN, visual network.

Correlation analysis results

The Spearman correlation analyses showed that the FC values in VN1 and VN2 were not significantly associated with either the duration of retinal detachment or the height of macular foveal detachment; specifically, the correlation between the VN1 FC and duration of detachment was ρ=−0.2012 (P=0.2070), and that between the VN2 FC and duration of detachment was ρ=−0.1847 (P=0.2477), while the correlation between the VN1 FC and foveal detachment height was ρ=0.0864 (P=0.6273), and that between VN2 FC and foveal detachment height was ρ=0.0264 (P=0.8820); detailed scatter plots illustrating these relationships are provided in Appendix 1.

Validation analysis

To verify whether our main findings were influenced by the preprocessing strategy, we conducted a supplementary analysis incorporating global signal regression. The results confirmed the stability of the observed effects, suggesting that the functional alterations identified in the RRD patients were not artifacts introduced by the preprocessing pipeline. For detailed results, see Appendix 2 (Figures S1,S2, Table S1).

Discussion

In this study, we integrated surface-based ICA and SBM techniques to systematically analyze cortical functional and structural changes in RRD patients, and applied SVM and SHAP methods to classify these changes. In terms of cortical function, the RRD patients demonstrated increased FC values in visual, cognitive, auditory, and motor networks, with notably significant increases in FNC between the VN and the SMN. In terms of cortical structure, the RRD group exhibited significant cortical thinning compared to the HC group in the lh_lateraloccipital, rh_pericalcarine, and rh_lateralorbitofrontal. Notably, the FC value of VN1 demonstrated the highest classification performance in both the SVM and SHAP models. These findings suggest that RRD patients experience significant alterations in cortical function and structure, particularly in networks related to visual and perceptual processing, and that there is a strong correlation between these changes and cortical thinning.

Cortical functional networks are groups of interconnected brain regions in the cerebral cortex that collaborate to support various cognitive and physiological functions (45). These networks are spatially and functionally integrated, enabling the brain to perform complex tasks such as perception, motor control, attention regulation, and higher cognitive functions. For instance, the VN includes regions involved in processing visual information, such as the primary visual cortex (V1) and higher-level visual areas. These areas receive, analyze, and interpret external visual stimuli (46,47). The DMN is active when the brain is at rest or engaged in self-reflection, supporting processes like introspection, self-awareness, and future planning (48-50). The AN involves the auditory cortex in the temporal lobes, which processes auditory information such as sound, speech, and music (51-53). The DAN regulates spatial attention and goal-directed behavior, helping individuals focus attention and respond to environmental stimuli (54-56). The SMN includes regions that control bodily movements and sensations, coordinating motor and sensory feedback (57,58). The coordinated function of these networks enables the brain to perform perceptual, motor, and higher cognitive tasks.

Previous studies have demonstrated that FC in networks can change in various diseases or pathological conditions. For example, Liu et al. (59) reported both increased FC in the VN and decreased FC in the DAN in patients with thyroid-associated ophthalmopathy. The enhanced VN connectivity may reflect a compensatory mechanism aimed at mitigating visual function decline by upregulating activity in relevant brain regions, whereas the reduced FC in the DAN could be linked to cognitive impairment in this patient population. Similarly, Jin et al. (60) found that patients with concomitant exotropia exhibited increased FC in the SMN and decreased FC in the AN. The former suggests a strong association between sensorimotor integration and the formation of stereoscopic vision, potentially explaining deficits in eye movement and depth perception, while the latter may reflect auditory dysfunction secondary to impaired stereovision. Moreover, a study on high myopia revealed increased FC in the DMN, implying that functional alterations in this network may underlie cognitive changes in affected individuals (61).

Other studies have also reported similar FC enhancements in visual or sensory-impaired populations that have been interpreted as reflecting network plasticity or functional compensation. For example, Iraji et al. (62) found that whole-brain FC increases in mild traumatic brain injury, which may be a compensatory response. Huang et al. (36) identified increased FNC between the VN and DMN subcomponents in patients with diabetic retinopathy, while Wang et al. (63) observed elevated connectivity between the SMN and DAN in patients with primary angle-closure glaucoma, potentially reflecting adaptation to visual information loss. Similarly, we identified significant increases in FC in multiple large-scale cortical networks, including the VN, DMN, AN, DAN, and SMN, in patients with RRD. Such widespread FC enhancement may indicate that the brain initiates large-scale network-level responses to acute visual disruption. Specifically, these changes may reflect early functional remodeling adaptations aimed at preserving or reallocating sensory and cognitive resources following retinal detachment. However, this interpretation remains speculative, as no behavioral or task-based validation was available in the current study. It is therefore unclear whether these alterations represent beneficial compensatory processes or non-specific reorganization. Future longitudinal and multimodal investigations incorporating neuropsychological assessments need to be conducted to elucidate the functional relevance of these connectivity changes and determine whether they reflect adaptive remodeling in the face of visual system damage.

FNC is used to assess the synchronization of activity between different brain regions, identifying abnormal connectivity patterns in various diseases. The SMN is crucial for regulating eye movements and encoding oculomotor activity (64), and is closely linked to the spontaneous activity of the V1 (46). This connection is essential for processing spatial visual information. Recent neuroimaging evidence suggests that alterations in FNC between the VN and SMN may occur in vision-related disorders, potentially reflecting compensatory reorganization during the early or subacute stages of visual disruption. For example, Liu et al. (65) observed significantly enhanced dynamic FNC between the VN and SMN in patients with diabetic retinopathy during connectivity state 1, interpreted as an adaptive response to peripheral retinal pathology. Similarly, Wang et al. (66) reported increased FC between the visual cortex and sensorimotor-related regions in patients with non-arteritic anterior ischemic optic neuropathy, suggesting that acute or short-term visual dysfunction may facilitate the transient reinforcement of VN-SMN interactions. These findings align with our results for RRD patients, and together, support the view that increased VN-SMN connectivity may reflect a stage-specific compensatory mechanism, distinct from the connectivity reductions commonly reported in long-term or congenital blindness.

CT is a key biological marker of brain structure and is widely used to investigate the pathological mechanisms of neurological diseases. In this study, we observed significant cortical thinning in RRD patients, particularly in the lh_lateraloccipital, rh_pericalcarine, and rh_lateralorbitofrontal regions. The lh_lateraloccipital and rh_pericalcarine regions, located in the occipital lobe, are crucial for visual processing, particularly in receiving and integrating visual stimuli from the retina (67,68). The rh_lateralorbitofrontal region, located in the frontal lobe, is involved in higher cognitive functions such as executive control, decision making, and emotional regulation (69). Previous studies have shown that visual impairment–related diseases often lead to changes in CT in the occipital and frontal lobes. For instance, Wu et al. (70) observed significant cortical thinning in the occipital visual cortex of patients with high myopia, which may result from functional alterations in visual processing areas. Yu et al. (71) also reported reduced CT in the visual cortex around the calcarine sulcus in patients with primary open-angle glaucoma, suggesting that damage to the lateral geniculate nucleus may underlie these structural changes. Children with anisometropic amblyopia also showed reduced occipital CT, likely due to impairments in VN projections and functional damage to the visual perception system (72). Conversely, studies on children with concomitant strabismus found decreased CT in the prefrontal area of the frontal lobe, possibly reflecting disruptions in the control of visual attention (73). Our findings align with these studies, further supporting CT changes as key markers of visual impairment and cognitive dysfunction. In RRD patients, cortical thinning in the lh_lateraloccipital and rh_pericalcarine regions may reflect adaptive changes due to retinal input loss, while thinning in the rh_lateralorbitofrontal region may indicate the cognitive impact of visual damage, particularly on emotional regulation and decision making.

To evaluate the diagnostic potential of cortical alterations, we applied machine learning methods, including SVM classification and SHAP analysis. The results indicated that the SVM + SHAP model based on FC values in the VN1 region achieved high classification accuracy in identifying RRD patients, suggesting its feasibility as a potential neuroimaging marker of RRD-related cortical dysfunction. Previous studies have also successfully implemented SVM combined with SHAP in various neuroimaging classification tasks, providing methodological support for our approach. For example, Vedaei et al. (74) applied this method to classify patients with mild traumatic brain injury using resting-state fMRI data, identifying degree centrality as the most discriminative feature. Similarly, studies on Parkinson’s disease (75), major depressive disorder (76), and end-stage renal disease (77) have employed SHAP-enhanced SVM models to identify clinically relevant imaging features, such as mean regional homogeneity in the superior frontal gyrus, resting-state FC, and network attributes. These findings confirm the interpretability and effectiveness of this approach in small- to moderate-sample neuroimaging contexts. Our findings are in line with these studies and further suggest that FC alterations in the VN1 region may serve as a candidate biomarker for detecting early cortical changes in RRD. Nonetheless, the generalizability of our model requires validation in larger and independent patient cohorts. In future work, we plan to incorporate external datasets for model validation and apply permutation testing and bootstrap resampling to evaluate its robustness and reproducibility. Overall, although preliminary, this study provides novel insights into early RRD detection and offers a promising framework for identifying high-risk individuals before clinical symptoms appear, thereby enabling earlier intervention and treatment.

Limitations

Despite the valuable insights provided by this study, several limitations should be acknowledged. First, the sample size of RRD patients was relatively small, which may limit the generalizability of our findings to a larger and more diverse population. Second, while the integration of surface-based ICA and SBM techniques allows for a comprehensive assessment of cortical changes, these methods inherently rely on certain assumptions, such as linearity in the relationship between brain structure and function, which may not fully capture the complexity of brain alterations in RRD. Third, our study is cross-sectional in nature and does not track cortical changes over time (e.g., pre- vs. post-operative), making it difficult to determine whether the observed alterations are pre-existing, reactive, or compensatory. To address this limitation, we intend to conduct a longitudinal follow-up study of RRD patients before and after retinal reattachment surgery, using multimodal neuroimaging at multiple timepoints to explore the trajectory and potential reversibility of cortical reorganization. Fourth, although we found significant correlations between functional and structural changes, further research is necessary to investigate the underlying neurobiological mechanisms driving these alterations. Fifth, while we observed significantly increased FC in multiple large-scale cortical networks, the current study did not include neuropsychological testing or task-based fMRI assessments. Therefore, the interpretation that these FC changes reflect functional compensation remains speculative and requires further behavioral validation in future studies. Sixth, we did not stratify patients based on macular status (i.e., macula-on vs. macula-off), which is a clinically important prognostic factor in RRD. Although we included macular foveal detachment height as a continuous variable, future studies should incorporate categorical macular involvement to better account for its potential influence on cortical alterations. Seventh, the present analysis did not include detailed retinal structural parameters, such as outer retinal thickness or ellipsoid zone integrity, which are known to reflect the severity of photoreceptor damage and are potentially associated with VN alterations. The absence of these variables was due to incomplete or inconsistent OCT data across patients. Future studies incorporating standardized retinal layer imaging protocols (e.g., outer nuclear layer thickness and ellipsoid zone status) need to be conducted to fully elucidate the relationship between retinal microstructure and central nervous system alterations in RRD. Eighth, due to the lack of visual acuity measurements in the HC group, we could not statistically control for potential differences in vision between groups. Although all the HCs were ophthalmologically screened and reported no vision complaints, future studies should include matched quantitative visual assessments to better account for this potential confounding variable. Finally, the use of SVM and SHAP methods, while effective in classification and interpretation, may be limited by the choice of features and the potential for overfitting, which could affect the robustness of the results.

Conclusions

Our study identifies significant cortical functional and structural changes in RRD patients, which may be linked to cognitive and visual impairments. Notably, alterations in the FC of the VN1 region may serve as a promising neuroimaging biomarker to differentiate between RRD patients and HCs. Our findings offer novel insights into the neuropathological mechanisms of RRD, and suggest potential biomarkers for early diagnosis and intervention.

Acknowledgments

We would like to thank the study participants for volunteering for this study. We gratefully acknowledge the contributions of personnel who implemented acquisition of the data.

Footnote

Reporting Checklist: The authors have completed the STROBE reporting checklist. Available at https://qims.amegroups.com/article/view/10.21037/qims-2025-628/rc

Data Sharing Statement: Available at https://qims.amegroups.com/article/view/10.21037/qims-2025-628/dss

Funding: This study was supported by the National Natural Science Foundation of China (No. 82160207), the Key Projects of Jiangxi Youth Science Fund (No. 20202ACBL216008), the Science and Technology Plan of the Jiangxi Provincial Health and Health Commission (No. 202130156), and the Graduate Student Research Project of Jiangxi Province (No. 2025KXJYS601).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://qims.amegroups.com/article/view/10.21037/qims-2025-628/coif). The authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study was approved by the Medical Ethics Committee of The First Affiliated Hospital of Nanchang University (approval No. IIT [2024] Ethics No. 790). All participants provided written informed consent. The study complied with the Declaration of Helsinki and its subsequent amendments.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Mitry D, Fleck BW, Wright AF, Campbell H, Charteris DG. Pathogenesis of rhegmatogenous retinal detachment: predisposing anatomy and cell biology. Retina 2010;30:1561-72. [Crossref] [PubMed]
2. Lake SR, Bottema MJ, Williams KA, Lange T, Reynolds KJ. Retinal Shape-Based Classification of Retinal Detachment and Posterior Vitreous Detachment Eyes. Ophthalmol Ther 2023;12:155-65. [Crossref] [PubMed]
3. Feltgen N, Walter P. Rhegmatogenous retinal detachment--an ophthalmologic emergency. Dtsch Arztebl Int 2014;111:12-21; quiz 22. [Crossref] [PubMed]
4. Ge JY, Teo ZL, Chee ML, Tham YC, Rim TH, Cheng CY, Wong TY, Wong EYM, Lee SY, Cheung N. International incidence and temporal trends for rhegmatogenous retinal detachment: A systematic review and meta-analysis. Surv Ophthalmol 2024;69:330-6. [Crossref] [PubMed]
5. Ozgür S, Esgin H. Macular function of successfully repaired macula-off retinal detachments. Retina 2007;27:358-64. [Crossref] [PubMed]
6. Mitry D, Awan MA, Borooah S, Syrogiannis A, Lim-Fat C, Campbell H, Wright AF, Fleck BW, Charteris DG, Yorston D, Singh J. Long-term visual acuity and the duration of macular detachment: findings from a prospective population-based study. Br J Ophthalmol 2013;97:149-52. [Crossref] [PubMed]
7. van de Put MA, Hoeksema L, Wanders W, Nolte IM, Hooymans JM, Los LI. Postoperative vision-related quality of life in macula-off rhegmatogenous retinal detachment patients and its relation to visual function. PLoS One 2014;9:e114489. [Crossref] [PubMed]
8. Vujosevic S, Parra MM, Hartnett ME, O'Toole L, Nuzzi A, Limoli C, Villani E, Nucci P. Optical coherence tomography as retinal imaging biomarker of neuroinflammation/neurodegeneration in systemic disorders in adults and children. Eye (Lond) 2023;37:203-19. [Crossref] [PubMed]
9. Mutlu U, Ikram MK, Roshchupkin GV, Bonnemaijer PWM, Colijn JM, Vingerling JR, Niessen WJ, Ikram MA, Klaver CCW, Vernooij MW. Thinner retinal layers are associated with changes in the visual pathway: A population-based study. Hum Brain Mapp 2018;39:4290-301. [Crossref] [PubMed]
10. Huang X, Li D, Li HJ, Zhong YL, Freeberg S, Bao J, Zeng XJ, Shao Y. Abnormal regional spontaneous neural activity in visual pathway in retinal detachment patients: a resting-state functional MRI study. Neuropsychiatr Dis Treat 2017;13:2849-54. [Crossref] [PubMed]
11. Kang HH, Shu YQ, Yang L, Zhu PW, Li D, Li QH, Min YL, Ye L, Zhou Q, Shao Y. Measuring abnormal intrinsic brain activities in patients with retinal detachment using amplitude of low-frequency fluctuation: a resting-state fMRI study. Int J Neurosci 2019;129:681-6. [Crossref] [PubMed]
12. Ji Y, Wang YY, Cheng Q, Fu WW, Shu BL, Wei B, Huang QY, Wu XR. Aberrant dynamic functional and effective connectivity changes of the primary visual cortex in patients with retinal detachment via machine learning. Neuroreport 2024;35:1071-81. [Crossref] [PubMed]
13. McKeown MJ, Makeig S, Brown GG, Jung TP, Kindermann SS, Bell AJ, Sejnowski TJ. Analysis of fMRI data by blind separation into independent spatial components. Hum Brain Mapp 1998;6:160-88. [Crossref] [PubMed]
14. McKeown MJ, Hansen LK, Sejnowsk TJ. Independent component analysis of functional MRI: what is signal and what is noise? Curr Opin Neurobiol 2003;13:620-9. [Crossref] [PubMed]
15. Beckmann CF, DeLuca M, Devlin JT, Smith SM. Investigations into resting-state connectivity using independent component analysis. Philos Trans R Soc Lond B Biol Sci 2005;360:1001-13. [Crossref] [PubMed]
16. Wang D, Qin W, Liu Y, Zhang Y, Jiang T, Yu C. Altered resting-state network connectivity in congenital blind. Hum Brain Mapp 2014;35:2573-81. [Crossref] [PubMed]
17. Tu Y, Fu Z, Mao C, Falahpour M, Gollub RL, Park J, Wilson G, Napadow V, Gerber J, Chan ST, Edwards RR, Kaptchuk TJ, Liu T, Calhoun V, Rosen B, Kong J. Distinct thalamocortical network dynamics are associated with the pathophysiology of chronic low back pain. Nat Commun 2020;11:3948. [Crossref] [PubMed]
18. Cai H, Wang C, Qian Y, Zhang S, Zhang C, Zhao W, Zhang T, Zhang B, Chen J, Liu S, Zhu J, Yu Y. Large-scale functional network connectivity mediate the associations of gut microbiota with sleep quality and executive functions. Hum Brain Mapp 2021;42:3088-101. [Crossref] [PubMed]
19. Roland PE, Zilles K. Structural divisions and functional fields in the human cerebral cortex. Brain Res Brain Res Rev 1998;26:87-105. [Crossref] [PubMed]
20. Tucholka A, Fritsch V, Poline JB, Thirion B. An empirical comparison of surface-based and volume-based group studies in neuroimaging. Neuroimage 2012;63:1443-53. [Crossref] [PubMed]
21. Coalson TS, Van Essen DC, Glasser MF. The impact of traditional neuroimaging methods on the spatial localization of cortical areas. Proc Natl Acad Sci U S A 2018;115:E6356-65. [Crossref] [PubMed]
22. Dahnke R, Yotter RA, Gaser C. Cortical thickness and central surface estimation. Neuroimage 2013;65:336-48. [Crossref] [PubMed]
23. Riccelli R, Toschi N, Nigro S, Terracciano A, Passamonti L. Surface-based morphometry reveals the neuroanatomical basis of the five-factor model of personality. Soc Cogn Affect Neurosci 2017;12:671-84. [Crossref] [PubMed]
24. Chklovskii DB, Mel BW, Svoboda K. Cortical rewiring and information storage. Nature 2004;431:782-8. [Crossref] [PubMed]
25. la Fougère C, Grant S, Kostikov A, Schirrmacher R, Gravel P, Schipper HM, Reader A, Evans A, Thiel A. Where in-vivo imaging meets cytoarchitectonics: the relationship between cortical thickness and neuronal density measured with high-resolution [18F]flumazenil-PET. Neuroimage 2011;56:951-60.
26. Clarkson MJ, Cardoso MJ, Ridgway GR, Modat M, Leung KK, Rohrer JD, Fox NC, Ourselin S. A comparison of voxel and surface based cortical thickness estimation methods. Neuroimage 2011;57:856-65. [Crossref] [PubMed]
27. Li B, Liu YX, Li HJ, Yuan Q, Zhu PW, Ye L, Su T, Shi WQ, Lin Q, Min YL, Xu XW, Shao Y. Reduced gray matter volume in patients with retinal detachment: evidence from a voxel-based morphometry study. Acta Radiol 2020;61:395-403. [Crossref] [PubMed]
28. Orrù G, Pettersson-Yeo W, Marquand AF, Sartori G, Mechelli A. Using Support Vector Machine to identify imaging biomarkers of neurological and psychiatric disease: a critical review. Neurosci Biobehav Rev 2012;36:1140-52. [Crossref] [PubMed]
29. Gaonkar B. T Shinohara R, Davatzikos C; Alzheimers Disease Neuroimaging Initiative. Interpreting support vector machine models for multivariate group wise analysis in neuroimaging. Med Image Anal 2015;24:190-204. [Crossref] [PubMed]
30. Jayasumana S, Hartley R, Salzmann M, Li H, Harandi M. Kernel Methods on Riemannian Manifolds with Gaussian RBF Kernels. IEEE Trans Pattern Anal Mach Intell 2015;37:2464-77. [Crossref] [PubMed]
31. Pereira F, Mitchell T, Botvinick M. Machine learning classifiers and fMRI: a tutorial overview. Neuroimage 2009;45:S199-209. [Crossref] [PubMed]
32. Wang Z. A hybrid SVM-GLM approach for fMRI data analysis. Neuroimage 2009;46:608-15. [Crossref] [PubMed]
33. Lundberg SM, Erion G, Chen H, DeGrave A, Prutkin JM, Nair B, Katz R, Himmelfarb J, Bansal N, Lee SI. From Local Explanations to Global Understanding with Explainable AI for Trees. Nat Mach Intell 2020;2:56-67. [Crossref] [PubMed]
34. Esteban O, Markiewicz CJ, Blair RW, Moodie CA, Isik AI, Erramuzpe A, Kent JD, Goncalves M, DuPre E, Snyder M, Oya H, Ghosh SS, Wright J, Durnez J, Poldrack RA, Gorgolewski KJ. fMRIPrep: a robust preprocessing pipeline for functional MRI. Nat Methods 2019;16:111-6. [Crossref] [PubMed]
35. Wang Y, Li TQ. Dimensionality of ICA in resting-state fMRI investigated by feature optimized classification of independent components with SVM. Front Hum Neurosci 2015;9:259. [Crossref] [PubMed]
36. Huang X, Tong Y, Qi CX, Dan HD, Deng QQ, Shen Y. Large-Scale Neuronal Network Dysfunction in Diabetic Retinopathy. Neural Plast 2020;2020:6872508. [Crossref] [PubMed]
37. Wang Y, Wang C, Miao P, Liu J, Wei Y, Wu L, Wang K, Cheng J. An imbalance between functional segregation and integration in patients with pontine stroke: A dynamic functional network connectivity study. Neuroimage Clin 2020;28:102507. [Crossref] [PubMed]
38. Zuo XN, Kelly C, Adelstein JS, Klein DF, Castellanos FX, Milham MP. Reliable intrinsic connectivity networks: test-retest evaluation using ICA and dual regression approach. Neuroimage 2010;49:2163-77. [Crossref] [PubMed]
39. Wang C, Cai H, Sun X, Si L, Zhang M, Xu Y, Qian Y, Zhu J. Large-Scale Internetwork Functional Connectivity Mediates the Relationship between Serum Triglyceride and Working Memory in Young Adulthood. Neural Plast 2020;2020:8894868. [Crossref] [PubMed]
40. Yotter RA, Dahnke R, Thompson PM, Gaser C. Topological correction of brain surface meshes using spherical harmonics. Hum Brain Mapp 2011;32:1109-24. [Crossref] [PubMed]
41. Goto M, Abe O, Aoki S, Hayashi N, Miyati T, Takao H, Iwatsubo T, Yamashita F, Matsuda H, Mori H, Kunimatsu A, Ino K, Yano K, Ohtomo KJapanese Alzheimer's Disease Neuroimaging Initiative. Diffeomorphic Anatomical Registration Through Exponentiated Lie Algebra provides reduced effect of scanner for cortex volumetry with atlas-based method in healthy subjects. Neuroradiology 2013;55:869-75. [Crossref] [PubMed]
42. Ashburner J. A fast diffeomorphic image registration algorithm. Neuroimage 2007;38:95-113. [Crossref] [PubMed]
43. Yotter RA, Thompson PM, Gaser C. Algorithms to improve the reparameterization of spherical mappings of brain surface meshes. J Neuroimaging 2011;21:e134-47. [Crossref] [PubMed]
44. Song S, Zhan Z, Long Z, Zhang J, Yao L. Comparative study of SVM methods combined with voxel selection for object category classification on fMRI data. PLoS One 2011;6:e17191. [Crossref] [PubMed]
45. Petersen SE, Sporns O. Brain Networks and Cognitive Architectures. Neuron 2015;88:207-19. [Crossref] [PubMed]
46. Wang K, Jiang T, Yu C, Tian L, Li J, Liu Y, Zhou Y, Xu L, Song M, Li K. Spontaneous activity associated with primary visual cortex: a resting-state FMRI study. Cereb Cortex 2008;18:697-704. [Crossref] [PubMed]
47. Bourne JA, Cichy RM, Kiorpes L, Morrone MC, Arcaro MJ, Nielsen KJ. Development of Higher-Level Vision: A Network Perspective. J Neurosci 2024;44:e1291242024. [Crossref] [PubMed]
48. Buckner RL, Andrews-Hanna JR, Schacter DL. The brain's default network: anatomy, function, and relevance to disease. Ann N Y Acad Sci 2008;1124:1-38. [Crossref] [PubMed]
49. Sestieri C, Corbetta M, Romani GL, Shulman GL. Episodic memory retrieval, parietal cortex, and the default mode network: functional and topographic analyses. J Neurosci 2011;31:4407-20. [Crossref] [PubMed]
50. Menon V. 20 years of the default mode network: A review and synthesis. Neuron 2023;111:2469-87. [Crossref] [PubMed]
51. Ahmad Z, Balsamo LM, Sachs BC, Xu B, Gaillard WD. Auditory comprehension of language in young children: neural networks identified with fMRI. Neurology 2003;60:1598-605. [Crossref] [PubMed]
52. Hall DA, Hart HC, Johnsrude IS. Relationships between human auditory cortical structure and function. Audiol Neurootol 2003;8:1-18. [Crossref] [PubMed]
53. Woods DL, Alain C. Functional imaging of human auditory cortex. Curr Opin Otolaryngol Head Neck Surg 2009;17:407-11. [Crossref] [PubMed]
54. Fox MD, Corbetta M, Snyder AZ, Vincent JL, Raichle ME. Spontaneous neuronal activity distinguishes human dorsal and ventral attention systems. Proc Natl Acad Sci U S A 2006;103:10046-51. [Crossref] [PubMed]
55. Vossel S, Geng JJ, Fink GR. Dorsal and ventral attention systems: distinct neural circuits but collaborative roles. Neuroscientist 2014;20:150-9. [Crossref] [PubMed]
56. Machner B, Braun L, Imholz J, Koch PJ, Münte TF, Helmchen C, Sprenger A. Resting-State Functional Connectivity in the Dorsal Attention Network Relates to Behavioral Performance in Spatial Attention Tasks and May Show Task-Related Adaptation. Front Hum Neurosci 2021;15:757128. [Crossref] [PubMed]
57. Caspers J, Rubbert C, Eickhoff SB, Hoffstaedter F, Südmeyer M, Hartmann CJ, Sigl B, Teichert N, Aissa J, Turowski B, Schnitzler A, Mathys C. Within- and across-network alterations of the sensorimotor network in Parkinson's disease. Neuroradiology 2021;63:2073-85. [Crossref] [PubMed]
58. Han Z, Liu T, Shi Z, Zhang J, Suo D, Wang L, Chen D, Wu J, Yan T. Investigating the heterogeneity within the somatosensory-motor network and its relationship with the attention and default systems. PNAS Nexus 2023;2:pgad276. [Crossref] [PubMed]
59. Liu H, Zhong YL, Huang X. Specific static and dynamic functional network connectivity changes in thyroid-associated ophthalmopathy and it predictive values using machine learning. Front Neurosci 2024;18:1429084. [Crossref] [PubMed]
60. Jin H, Chen RB, Zhong YL, Lai PH, Huang X. Effect of Impaired Stereoscopic Vision on Large-Scale Resting-State Functional Network Connectivity in Comitant Exotropia Patients. Front Neurosci 2022;16:833937. [Crossref] [PubMed]
61. Zhang XW, Dai RP, Cheng GW, Zhang WH, Long Q. Altered amplitude of low-frequency fluctuations and default mode network connectivity in high myopia: a resting-state fMRI study. Int J Ophthalmol 2020;13:1629-36. [Crossref] [PubMed]
62. Iraji A, Chen H, Wiseman N, Welch RD, O'Neil BJ, Haacke EM, Liu T, Kou Z. Compensation through Functional Hyperconnectivity: A Longitudinal Connectome Assessment of Mild Traumatic Brain Injury. Neural Plast 2016;2016:4072402. [Crossref] [PubMed]
63. Wang Y, Shu Y, Cai G, Guo Y, Gao J, Chen Y, Lv L, Zeng X. Altered static and dynamic functional network connectivity in primary angle-closure glaucoma patients. Sci Rep 2024;14:11682. [Crossref] [PubMed]
64. Blanke O, Spinelli L, Thut G, Michel CM, Perrig S, Landis T, Seeck M. Location of the human frontal eye field as defined by electrical cortical stimulation: anatomical, functional and electrophysiological characteristics. Neuroreport 2000;11:1907-13. [Crossref] [PubMed]
65. Liu H, Gu ZX, Li XT, Huang X. Extensive Dynamic Functional Network Connectivity Alterations in Diabetic Retinopathy Among Patients with Type 2 Diabetes. Diabetes Metab Syndr Obes 2025;18:1823-35. [Crossref] [PubMed]
66. Wang H, Yan X, Ma N, Wu Q, Zhang Q, Zhou J, Guo P. Association of aberrant brain network connectivity with visual dysfunction in patients with nonarteritic anterior ischemic optic neuropathy: a pilot study. Quant Imaging Med Surg 2025;15:2362-75. [Crossref] [PubMed]
67. Grill-Spector K, Malach R. The human visual cortex. Annu Rev Neurosci 2004;27:649-77. [Crossref] [PubMed]
68. Ancelin ML, Carrière I, Artero S, Maller J, Meslin C, Ritchie K, Ryan J, Chaudieu I. Lifetime major depression and grey-matter volume. J Psychiatry Neurosci 2019;44:45-53. [Crossref] [PubMed]
69. Bechara A, Damasio H, Damasio AR. Emotion, decision making and the orbitofrontal cortex. Cereb Cortex 2000;10:295-307. [Crossref] [PubMed]
70. Wu YJ, Wu N, Huang X, Rao J, Yan L, Shi L, Huang H, Li SY, Zhou FQ, Wu XR. Evidence of cortical thickness reduction and disconnection in high myopia. Sci Rep 2020;10:16239. [Crossref] [PubMed]
71. Yu L, Xie B, Yin X, Liang M, Evans AC, Wang J, Dai C. Reduced cortical thickness in primary open-angle glaucoma and its relationship to the retinal nerve fiber layer thickness. PLoS One 2013;8:e73208. [Crossref] [PubMed]
72. Qi S, Mu YF, Cui LB, Li R, Shi M, Liu Y, Xu JQ, Zhang J, Yang J, Yin H. Association of Optic Radiation Integrity with Cortical Thickness in Children with Anisometropic Amblyopia. Neurosci Bull 2016;32:51-60. [Crossref] [PubMed]
73. Yin X, Chen L, Ma M, Zhang H, Gao M, Wu X, Li Y. Altered Brain Structure and Spontaneous Functional Activity in Children With Concomitant Strabismus. Front Hum Neurosci 2021;15:777762. [Crossref] [PubMed]
74. Vedaei F, Mashhadi N, Zabrecky G, Monti D, Navarreto E, Hriso C, Wintering N, Newberg AB, Mohamed FB. Identification of chronic mild traumatic brain injury using resting state functional MRI and machine learning techniques. Front Neurosci 2022;16:1099560. [Crossref] [PubMed]
75. Pang H, Yu Z, Yu H, Cao J, Li Y, Guo M, Cao C, Fan G. Use of machine learning method on automatic classification of motor subtype of Parkinson's disease based on multilevel indices of rs-fMRI. Parkinsonism Relat Disord 2021;90:65-72. [Crossref] [PubMed]
76. Dai P, Xiong T, Zhou X, Ou Y, Li Y, Kui X, Chen Z, Zou B, Li W, Huang ZThe Rest-Meta-Mdd Consortium. The alterations of brain functional connectivity networks in major depressive disorder detected by machine learning through multisite rs-fMRI data. Behav Brain Res 2022;435:114058. [Crossref] [PubMed]
77. Yu Z, Du Y, Pang H, Li X, Liu Y, Bu S, Wang J, Zhao M, Ren Z, Li X, Yao L. Effects of Renal Function on the Multimodal Brain Networks Affecting Mild Cognitive Impairment Converters in End-Stage Renal Disease. Acad Radiol 2025;32:4029-40. [Crossref] [PubMed]