Utilizing multivariate pattern analysis to uncover the neurobiological underpinnings of communication disorders in children with bilateral spastic cerebral palsy: insights from morphological and structural connectivity changes

Introduction

Cerebral palsy (CP) is a prevalent physical disability affecting 2–3 per 1,000 live births (1), with bilateral spastic cerebral palsy (BSCP) being the predominant type among children (2). Communication impairments, including receptive and expressive language, are comorbidities in patients with CP, with prevalence rates ranging from 46–78% (3,4). Effective communication is a critical enabler of daily engagement and social participation (5,6). For children with BSCP, communication impairments often hinder self-development, learning, and social integration, resulting in severely restricted activity participation (7).

Given these significant challenges, there is an urgent need for a quantitative and individualized diagnostic model to assess and predict communication abilities in children with BSCP. Current clinical scales are limited, as many standardized tools rely on behavioral responses that young children with CP may not reliably perform (8,9). Non-standardized tests, such as parent-completed measures and grad-level classification systems, are unlikely to capture the full scope of necessary information to effectively track impairment-related changes (10). Thus, more objective and reliable methods are essential to accurately assess and track communication impairments in this population. Magnetic resonance imaging (MRI) has emerged as a valuable tool for investigating the neurobiological underpinnings of communication impairments in children with CP. Structural and functional MRI studies have revealed alterations in gray matter volumes and disrupted white matter connectivity associated with communication impairments (11). Notably, the severity of periventricular white matter lesions (PWML) correlates with worse communication performance, while children with BSCP exhibiting more severe lesions in language-related pathways tend to show greater deficits in communication skills (4,12). However, these studies have predominantly reported group-level trends and qualitative or semi-quantitative findings, highlighting the need for a quantitative approach tailored to individual patients.

Multivariate pattern analysis (MVPA) has become a widely used method in neuroscience. It provides a powerful framework for building individualized diagnostic models by integrating multiple imaging features simultaneously (13). MVPA can reveal complex brain network patterns related to communication impairment by comprehensively analyzing brain morphology and connectivity data. MVPA promotes the establishment of personalized diagnosis and prediction models, and its efficiency has been proven in the research of predicting the forecasting language in post-stroke patients and motor outcomes in neonates with arterial ischemic stroke (14-16). However, its application to predicting communication impairments in children with BSCP remains unexplored. This study seeks to fill this gap by employing MVPA to investigate brain morphological and white matter connectivity changes associated with communication impairments in children with BSCP. By selecting cortical and subcortical morphological features and white matter connectivity as MVPA model parameters, we aim to capture key neurobiological correlates of communication performance. Decreased gray matter volumes and disrupted white matter connections are associated with impaired communication function (17). Thus, these gray matter morphology and white matter connections index, derived from three-dimensional T1-weighted imaging (3D-T₁WI) and diffusion tensor imaging (DTI), provide an appropriate basis for MVPA-based prediction models. To further enhance prediction accuracy, support vector classification (SVC) is employed to classify communication abilities, as SVC has been proven effective for small-sample neuroimaging datasets (18). This combination of advanced imaging features and machine learning-based modeling not only improves diagnostic accuracy but also provides new insights into the neurobiological basis of communication impairments in BSCP.

This study tests two pivotal hypotheses: (I) the integration of multivariate imaging phenotyping and connectomics-derived network features will outperform unimodal approaches in diagnostic accuracy for communication impairments; (II) patient-specific predictive modeling can transcend the ecological validity constraints of group-level neurodevelopmental analyses while generating objective multimodal biomarkers to guide individualized rehabilitation. By systematically evaluating these hypotheses through MVPA implementation in children with BSCP, this investigation aims to establish a novel framework for precision diagnostics in neurodevelopmental communication disorders. We present this article in accordance with the STROBE reporting checklist (available at https://qims.amegroups.com/article/view/10.21037/qims-2025-861/rc).

Methods

Participants and clinical testing

The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Ethics Committee for biomedical research of the Affiliated Hospital of Zunyi Medical University (No. KLLY-2021-081) and written informed consent was obtained from the guardians of children with BSCP.

A total of 28 children diagnosed with BSCP were recruited between April 2019 and March 2022. Inclusion criteria were: (I) a confirmed diagnosis of BSCP; (II) age range from 4 to 16 years; (III) presence of PWML as confirmed by brain MRI. Exclusion criteria included: (I) history of other psychiatric or neurological disorders; (II) severe auditory or visual impairments; (III) inability to undergo 3D-T₁ or DTI scans due to medical or logistical reasons. For the control group, thirty-one age and sex-matched typically developing children were recruited. The exclusion criteria for this group were refined to ensure a clear differentiation from the patient group, including: (I) any MRI brain abnormalities; (II) any neurological conditions or developmental delays; (III) recent use of central nervous system affecting medication.

The standardized assessment process of Communication Function Classification System (CFCS), as described in our previously published protocol (19), was performed for children with BSCP. The Speech comprehension performance of children with BSCP and the healthy control (HC) group was evaluated using either the Wechsler Intelligence Scale for Children (fourth edition) or the Wechsler Preschool and Primary Scale of Intelligence (fourth edition) by experienced pediatricians. The verbal comprehension index (VCI) was extracted from each participant’s intelligence scale to assess their verbal comprehension performance.

MR acquisition and postprocessing

MRIs were acquired using a 3.0T scanner (Signa HDXT; GE Healthcare, Milwaukee, Wisconsin, USA) with an 8-channel head coil. The following MRI sequences were used—3D-T₁WI: repetition time (TR) =7.8 ms, echo time (TE) =3.0 ms, inversion time (TI) =450 ms, thickness =1.0 mm, spacing =0 mm, field of view (FOV) =256 mm × 256 mm, matrix size =256×256, flip angle =15°; T2-fluid-attenuated inversion recovery (T2-FLAIR): TR =7,500 ms, TE =140 ms, TI =2,100 ms, thickness =3 mm, spacing =1.5 mm, FOV =240 mm × 240 mm, matrix size =48×48, flip angle =90°; DTI: single-shot spin-echo (EPI), TR =12,500 ms, TE =85.8 ms, b =0, 1,000 mm²/s, flip angle =90°, 64 diffusion directions, thickness =2.5 mm, spacing =0 mm, FOV =240 mm × 240 mm, matrix size =96×96.

Cortical reconstruction and volumetric segmentation were performed on the 3DT₁-weighted images using Freesurfer 5.30 (http://surfer.nmr.mgh.harvard.edu). The Destrieux atlas was used to segment cortical and subcortical regions, with each voxel receiving an anatomical label. Manual corrections were performed where necessary, followed by visual verification by trained raters to ensure precision. Cortical thickness, area, volume, and subcortical volumes were extracted for each region.

DTI were processed using MRtrix3 (http://www.mrtrix.org/) software. According to our previous quality control method for diffusion data (20), participants with insufficient gradient directions or motion artifacts were excluded. Taking b0 image as the reference, images in different gradient directions were checked slice by slice using the two-dimensional local Pearson correlation coefficient. The threshold for motion artifact detection and exclusion was set to be the SD by a factor of 3 from the average of correlation coefficients. At least 2 nonzero b values and 30 gradient directions per nonzero b value should be retained to estimate the fractional anisotropy (FA) after the artifacts removal and removal of directions with excessive head movement. Motion and distortion correction were then performed on the remaining participants (20). Tensors were fitted, FA and fiber numbers (FN) values were computed. The response function of DWI data was estimated and fiber orientation distributions were derived using constrained spherical deconvolutions (21). Whole-brain probabilistic tractography was generated, and spherical-deconvolution informed filtering of tractograms (SIFT) was applied (22).

White matter connectome construction

The Destrieux atlas in native T₁ space was co-registered into DTI native space to define nodes of the network by using the bbregister tool from Freesurfer software (http://surfer.nmr.mgh.harvard.edu/fswiki/bbregister). The native space atlases were visually inspected pre- and post-registration. The streamlines outputted by SIFT were parcellated into a set of 148 regions of the Destrieux atlas for each participant using tck2connectome command (23). Then, two types of connectivity maps were created based on the number of streamlines and their mean FA for each edge (connection between nodes). The Connectivity map based on the number of streamlines was further processed by dividing the number of streamlines connecting two nodes by the sum of their volumes. Only edges that existed in 100% of the HC group were included in the subsequent analyses for both types of connectivity maps.

MVPA

An SVC algorithm was adopted to assess communication performance of children with BSCP using the features obtained from 3D-T₁WI and DTI image processing. Children with BSCP classified as CFCS I level were defined as not having communication impairment, while CFCS II–V levels were defined as having communication impairment. Crucially, for the BSCP-only classification models, feature selection was performed exclusively within the BSCP subgroup using t-tests to identify significant differences between impaired (CFCS II–V) and non-impaired (CFCS I) children; HC data played no role in this feature selection process.

The performance of the SVC model was evaluated using a leave-one-out cross-validation (LOOCV) procedure. Since the number of preselected features was still too large compared to the sample size, further feature selection was conducted using t-test in the training set. The training set was utilized to train a linear SVC model for classification, and a 5-fold cross-validation was employed to determine the optimal hyperparameter c. The optimal parameter c was applied to the training set to obtain the SVC model, which was then used in the validation set to predict the label of the remaining one BSCP child. The LOOCV procedure was repeated for all the children with BSCP, and each BSCP child received a predicted label. The accuracy, sensitivity, and specificity of the classifier were calculated based on the predicted labels and true labels of the children with BSCP.

The children with BSCP assigned to the training dataset differed for each iteration of LOOCV. Meanwhile, feature selection was only performed based on the training dataset, resulting in different features used for classification in each iteration. Consensus features were defined as features that were selected in more than 1/2 iterations in all LOOCV procedures. Finally, after disrupting the labels of children with BSCP, 1,000 permutation tests were used to assess whether the resulting model classification accuracy was statistically significant.

Statistics

The demographic and clinical differences between the BSCP and control groups were analyzed by Chi-squared test for categorical variables and the t-test for continuous variables. For cortical parameters, a two-sample t-test was used to compare the differences between BSCP children and control children, with age and gender included as covariates. To account for multiple comparisons, the false discovery rate (FDR) correction was applied. For white matter connectivity indicators, a two-sample t-test was employed to compare the differences between children with BSCP and the control group. The t-threshold was set to 2.5, and Network-Based Statistic method was used for multiple comparisons and correction. Differences in cortical morphological and white matter connectivity parameters were extracted from children with BSCP and were then subjected to partial Spearman correlation analysis with CFCS and VCI, controlling for age and gender as covariates. P values less than 0.05 were considered statistically significant.

Results

Demographic and clinical information

The study comprised 28 BSCP (14 male/14 female, 4–15 years) and 31 HC children (17 male/14 female, 4–13 years). Incomplete VCI assessments were noted for 3 BSCP and 6 HC children due to cooperation difficulties. No significant differences were found between the two groups in terms of age or gender. However, the VCI scores of children with BSCP were significantly lower compared to the HC group (71.3±20.7 vs. 103.3±13.7, P<0.001, Table 1).

Morphological differences of cortical and subcortical structures and their correlation with communication function

BSCP children showed reduced cortical surface areas in bilateral frontal lobes and left temporal lobes (FDR correction, P<0.05; Figure 1A, Table S1). Significant gray matter volume reductions were found in multiple regions, including the bilateral frontal lobes, left middle frontal gyrus, orbital frontal gyrus, anterior part of the middle cingulate gyrus, lower part of the corpus callosum, and upper part of the superior temporal sulcus (FDR correction, P<0.05; Figure 1B, Table S2). In subcortical regions, volumes of the bilateral thalamus, globus pallidus, lenticular nucleus, hippocampus, amygdala, left putamen, and corpus callosum were significantly reduced in the BSCP group (FDR correction, P<0.05; Figure 1C, Table S3).

Figure 1 Group differences in cortical and subcortical structures between BSCP children and healthy controls. (A) Regions with significantly reduced cortical surface area in children with BSCP (FDR corrected, P<0.05). (B) Regions with significantly reduced cortical gray matter volume in BSCP children (FDR corrected, P<0.05). (C) Subcortical structures with significantly reduced volumes in BSCP children (FDR corrected, P<0.05). BSCP, bilateral spastic cerebral palsy; FDR, false discovery rate.

In particular, the surface areas of the left middle temporal gyrus, middle frontal sulcus, and right occipital pole cortex were significantly negatively correlated with CFCS [FDR correction, rank correlation coefficient (rho) =−0.39, −0.41, −0.45; P<0.05] (Figure 2A-2C). Additionally, gray matter volumes in the right transversus frontal gyrus/sulcus, middle frontal gyrus, and sup frontal gyrus were negatively correlated with CFCS (FDR correction, rho =−0.39, −0.40, −0.54; P<0.05) and gray matter volumes in the left middle frontal sulcus was positively correlated with VCI (FDR correction, rho =0.47, P=0.426) (Figure 2D-2G). Negative correlations with CFCS were observed in the bilateral thalamus, hippocampus, amygdala and left putamen, globus pallidus (FDR correction, rho =−0.63 to −0.39; P<0.05) (Figure 2H-2O).

Figure 2 Morphological differences of cortical and subcortical structures and their correlation with communication function. Correlations between morphological measures and communication function in BSCP children. (A-C) Negative correlations between cortical surface area of specific regions (left middle temporal gyrus, left middle frontal sulcus, right occipital pole) and CFCS levels. (D-G) Correlations between cortical gray matter volume of specific regions (right lateral frontal gyrus/sulcus, right middle frontal gyrus, right superior frontal gyrus, left middle frontal sulcus) and communication function measures (CFCS and VCI). (H-O) Negative correlations between subcortical structure volumes (bilateral thalamus, hippocampus, amygdala, left putamen, left globus pallidus) and CFCS levels. BSCP, bilateral spastic cerebral palsy; CFCS, Communication Function Classification System; rho, rank correlation coefficient; VCI, verbal comprehension index.

Structure connections changes and their correlation with communication function

After excluding children with incomplete or low-quality DTI data (5 HC and 3 BSCP children), 26 BSCP children and 25 HC children were included in the structural connectivity analysis. The BSCP group showed widespread reductions in FA values [network-based statistic (NBS) correction, P<0.05; Figure 3A], along with decreased FN (NBS correction, P<0.05; Figure 3B) between the bilateral frontoparietal lobes, temporal lobe, sensorimotor regions, and cingulate gyrus. The FA values of white matter tracts in the bilateral frontal lobes, sensorimotor areas, deep gray matter regions, inter-cingulate connections, and internal commissure were negatively correlated with CFCS (Figure 3C, Table S4). Additionally, positive correlations with VCI were found for FA values connecting the sensorimotor areas, frontal-parietal lobes, and cingulate gyrus. In contrast, negative correlations with VCI were found in FA values connecting the paracentral gyrus/sulcus, right temporal pole, and insular lobe (Figure 3D, Table S5). No significant correlations were found between FN and either CFCS or VCI.

Figure 3 Structural connectivity alterations and correlations with communication function. (A) Significantly reduced FA values in BSCP children compared to HC (P<0.05). (B) Significantly reduced FN in BSCP children compared to HC (P<0.05). (C) Negative correlations between FA values of white matter tracts and CFCS levels in BSCP children. (D) Correlations between FA values and VCI in BSCP children; red connections represent positive correlations and blue connections represent negative correlations. BSCP, bilateral spastic cerebral palsy; CFCS, Communication Function Classification System; FA, fractional anisotropy; FN, fiber numbers; HC, healthy controls; VCI, verbal comprehension index.

Models for the prediction of communication ability

As shown in Table 2, the linear SVC classifier using cortical surface area features achieved an accuracy of 78.5% (60.0% sensitivity, 88.8% specificity, P=0.015) with the top 15% ranked surface area features (Figure S1). Similarly, the cortical volume model achieved an accuracy of 71.4% (50.0% sensitivity, 77.7% specificity, P=0.08) using the top 25% ranked volume features (Figure S2). For the subcortical structure volume model, the SVC classifier reached an accuracy of 71.4% (70.0% sensitivity, 77.7% specificity, P=0.09) by selecting the top 10% ranked subcortical features (Figure S3). The FA-based model achieved an accuracy of 76.9% (90.0% sensitivity, 75.0% specificity, P=0.037) using the top 15 functional connections (Figure S4). The FN-based model achieved an accuracy of 73.0% (80.0% sensitivity, 81.2% specificity, P=0.04) using the top 60 ranked FN features (Figure S5). To enhance classification performance, we constructed a combined model using the same LOOCV method, based on the consistent discriminative features from the significant models (cortical surface area, FA, and FN models with P<0.05). This combined model achieved the highest accuracy of 80.7% and an area under the curve (AUC) of 0.88, indicating strong classification power (Table 2).

Consistent features

In the cortical surface area model, the consistent features across LOOCV iterations were located in the left middle frontal gyrus, left middle frontal sulcus, left middle temporal gyrus, left sulcus intermedius primus, right superior frontal gyrus, right transverse frontal pole gyrus and gyrus, right middle occipital and lunatus gyrus, right occipital pole and right cuneus (Figure 4). In the FA model, consistent features were observed within the left inferior frontal gyrus, within the bilateral sensorimotor regions, and between the bilateral precuneus, left marginal sulcus of the cingulate gyrus, inferior temporal gyrus, medial occipital-temporal sulcus, lingual sulcus, and pericallosal sulcus (Figure 5A). In the FN model, consistent features were found between and within the bilateral frontal lobes, sensorimotor regions, deep gray matter, superior temporal gyrus, pre-wedge gyrus, cingulate gyrus, and parieto-occipital regions (Figure 5B).

Figure 4 Consistent discriminative cortical surface area features identified in the classification model. L, left; R, right.

Figure 5 Consistent discriminative structural connectivity features identified in FA and FN models. (A) Consistent FA connectivity features were identified in the FA-based model. (B) Consistent FN connectivity features were identified in the FN-based model. FA, fractional anisotropy; FN, fiber numbers.

Discussion

This study demonstrates that reduced cortical surface areas and gray matter volumes, along with widespread white matter connectivity abnormalities, are associated with communication impairments in children with BSCP. These impairments are linked to morphological alterations and white matter disconnections within the bilateral frontal lobes, sensorimotor regions, and temporal and occipital lobes. MVPA of brain morphology and structural connectivity provides an effective means for predicting communication impairments in children with BSCP. Our findings validate the use of SVC-based models combining gray matter and white matter features for individualized diagnosis of communication impairments.

MVPA

This study confirms the utility of MVPA in classifying communication impairments by integrating brain morphology and structural connectivity. The diagnostic power of white matter-based models exceeded that of gray matter-based models, likely due to PWML being primarily characterized by white matter damage. White matter connectivity metrics, such as FA values and FN, better capture the underlying brain injury patterns compared to morphological features derived from 3D-T₁WI. Compared with traditional lesion-symptom mapping methods and prior predictive models based on univariate lesion characteristics or unimodal approaches in CP research (24,25), this study extends current knowledge: Multivariate patterns extracted from routine clinical MRI (3D-T₁WI + DTI) may have advantages in predicting specific functional outcomes (communication impairments); Providing initial evidence for individualized classification within the clinically critical BSCP subgroup, potentially advancing beyond group-level associations toward developing personalized assessment frameworks.

The highest AUC value was achieved by the combined diagnostic model, which integrates gray matter structure and white matter connectivity. This suggests that combining multimodal MRI parameters provides a more comprehensive understanding of communication impairments. Previous neuroimaging studies have primarily relied on univariate voxel-wise analyses, which consider each voxel as an independent unit and often fail to detect distributed effects across brain regions (26). In contrast, MVPA leverages spatially distributed patterns across multiple brain regions, enhancing the ability to identify complex injury patterns and predict individual outcomes (27). Our findings demonstrate the potential of MVPA to improve individual-level diagnostic predictions for communication impairments in children with BSCP.

Brain morphological features associated with communication impairments

The frontal and temporal lobes play a critical role in language processing (28,29). Our findings indicate that the left middle frontal gyrus is essential for phonetic and semantic processing and the right middle frontal gyrus are key regions associated with communication impairments. Reduced surface area of the left middle frontal gyrus and volume reduction in the right middle frontal gyrus were both negatively correlated with CFCS scores, indicating their involvement in communication deficits. Additionally, the volume of the left frontal sulcus was positively correlated with VCI scores, further highlighting the relationship between frontal lobe morphology and communication abilities. The left middle temporal gyrus is a crucial part of the semantic cognitive processing network (30,31). Our study found a reduction in the surface area of the left middle temporal gyrus, which was negatively correlated with CFCS. This aligns with studies of aphasia, where semantic processing tasks activate the bilateral middle temporal gyri (32).

Previous research shows that children with language disorders exhibit reduced neural activity in the left middle temporal gyrus during phonetic learning tasks (33). We speculate that atrophy in the left middle temporal gyrus disrupts semantic processing, contributing to communication impairments in children with BSCP.

Subcortical structures such as the hippocampus and amygdala also play a role in language processing. The bilateral hippocampal volume was negatively correlated with CFCS, and the left hippocampus distinguished between children with and without communication impairments. In addition to memory, the hippocampus is involved in semantic and phonetic fluency (34). Similarly, smaller amygdala volumes were associated with poorer communication performance. These findings are consistent with studies in children with autism, where amygdala size correlates with social communication abilities (35). Furthermore, reduced thalamus, putamen, and globus pallidus volumes were correlated with communication impairments. These findings suggest that deep gray matter damage contributes to the severity of communication impairments, reinforcing the importance of cortical and subcortical integrity in language development (4,12,36).

Brain connections changes associated with communication impairments

FA values in the left inferior frontal gyrus, sensorimotor regions, and bilateral pre-cuneiform gyrus-cingulate connections were negatively correlated with CFCS and positively correlated with VCI. Language production requires coordination across multiple brain regions, integrating auditory inputs with motor outputs. Damage to Broca’s area, the motor language center, can lead to expressive aphasia, including dysarthria and non-fluent speech (37). Our study suggests that information separation and integration dysfunction within Broca’s area contributes to communication impairments in BSCP children (38). The sensorimotor network plays a crucial role in speech production, with changes in white matter fibers within the sensorimotor region impacting motor aspects of speech. The FA values of these fiber bundles help distinguish children with communication impairments from those without and are negatively correlated with CFCS. We speculate that abnormal integration within the sensorimotor network disrupts speech production, contributing to communication impairments. The speech control network involves specific brain regions, including the prefrontal cortex, inferior parietal lobule, and precuneus (39). Our study found that FN values between the frontal, parietal, and cingulate regions were effective in distinguishing communication impairments. This suggests that communication impairments may result from dysfunctional large-scale network coordination. Abnormalities in these networks may activate multimodal integration regions abnormally, further disrupting the coordination required for fluent communication. Sensorimotor center uniquely activates and multimodal integrates cortical regions such as the prefrontal cortex, the lower parietal cortex, the precuneus lobe during the speech production process (39,40). This indicates that the speech control network also needs the coordination work of specific modules to collectively constitute the speech function connection structure. In our study FA value of bilateral pre-cuneiform gyrus-cingulate connection, FN value of frontal lobe-cingulate gyrus-parietal lobe connection and FN value between deep gray matter nuclei in BSCP children are helpful to distinguish whether communication impairment is combined, indicating that communication impairment in BSCP children may be associated with large-scale speech network. We speculated that abnormal network or internal connections in BSCP children might activate abnormal multimodal integration cortex regions, leading to abnormal coordination of large-scale brain networks, and resulting in communication impairments.

Limitations

The small sample size constrains model stability and increases overfitting risk in high-dimensional analyses. While we employed LOOCV to mitigate this, its diagnostic accuracy findings must be interpreted as preliminary due to inherent small-sample limitations. Dichotomizing the CFCS levels (I vs. II–V) simplifies a functional continuum and may obscure subtle differences between levels II and III. Future research should further explore the correlation mechanisms between brain morphology and white matter connectivity, and conduct external validation to optimize model applicability. Although the model constructed in this study has good performance, it lacks external verification. Additional data are needed to optimize the model to improve its applicability.

Conclusions

Morphological alterations in the brain structure and the variations in white matter connectivity are associated with communication impairments in BSCP children. MVPA-based models offer a promising approach for developing individualized predictive frameworks for these impairments. Our findings offer new insights into the neurobiological mechanisms underlying communication difficulties in children with BSCP and underscore the importance of multimodal MRI in advancing our understanding of these impairments.

Acknowledgments

We extend our sincere gratitude to all participating children and their families for their invaluable contributions to this study.

Footnote

Reporting Checklist: The authors have completed the STROBE reporting checklist. Available at https://qims.amegroups.com/article/view/10.21037/qims-2025-861/rc

Data Sharing Statement: Available at https://qims.amegroups.com/article/view/10.21037/qims-2025-861/dss

Funding: This work was supported by the Young Outstanding Scientific and Technological Talent of Guizhou Province (grant No. Qiankehepingtairencai [2021] 5620), Key Basic Research Program of Guizhou Province (grant No. Qiankehejichu-ZK [2022] zhongdian 051), Talent Program for Future Famous Clinical Doctors of Zunyi Medical University (grant No. rc220211205), Innovation Talent team of Science and Technology in Zunyi City (grant No. Zunshikerencai [2024] 5), Chongqing Medical Scientific Research Project (Joint project of Chongqing Health Commission and Science and Technology Bureau) (2026MSXM132), Chronic Disease Management Research Project of National Health Commission Capacity Building and Continuing Education Center (No.GWJJMB202510023052), Open subject of Chongqing Key Laboratory of Emergency Medicine (No. 2023KFKT04), Wuxi Municipal Health Commission’s “Double Hundred” Medical Health Young Elite Talent Project (grant No. BJ2023088) and the Wuxi Municipal Science and Technology Bureau’s Medical and Health Tackling Project (grant No. Y20232012).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://qims.amegroups.com/article/view/10.21037/qims-2025-861/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 conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Ethics Committee for biomedical research of the Affiliated Hospital of Zunyi Medical University (No. KLLY-2021-081) and written informed consent was obtained from the guardians of children with bilateral spastic cerebral palsy (BSCP).

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. Patel DR, Neelakantan M, Pandher K, Merrick J. Cerebral palsy in children: a clinical overview. Transl Pediatr 2020;9:S125-35. [Crossref] [PubMed]
2. Barnett R. Cerebral palsy. Lancet 2018;392:545. [Crossref] [PubMed]
3. Coleman A, Weir K, Ware RS, Boyd R. Predicting functional communication ability in children with cerebral palsy at school entry. Dev Med Child Neurol 2015;57:279-85. [Crossref] [PubMed]
4. Coleman A, Fiori S, Weir KA, Ware RS, Boyd RN. Relationship between brain lesion characteristics and communication in preschool children with cerebral palsy. Res Dev Disabil 2016;58:55-64. [Crossref] [PubMed]
5. Kristoffersson E, Dahlgren Sandberg A, Holck P. Communication ability and communication methods in children with cerebral palsy. Dev Med Child Neurol 2020;62:933-8. [Crossref] [PubMed]
6. Vaillant E, Geytenbeek JJM, Jansma EP, Oostrom KJ, Vermeulen RJ, Buizer AI. Factors associated with spoken language comprehension in children with cerebral palsy: a systematic review. Dev Med Child Neurol 2020;62:1363-73. [Crossref] [PubMed]
7. Williams PG, Lerner MACouncil on Early Childhood. Council on School Health. School Readiness. Pediatrics 2019;144:e20191766. [Crossref] [PubMed]
8. Pennington L, Dave M, Rudd J, Hidecker MJC, Caynes K, Pearce MS. Communication disorders in young children with cerebral palsy. Dev Med Child Neurol 2020;62:1161-9. [Crossref] [PubMed]
9. Tan SS, van Gorp M, Voorman JM, Geytenbeek JJ, Reinders-Messelink HA, Ketelaar M, Dallmeijer AJ, Roebroeck MEPerrin-Decade study group. Development curves of communication and social interaction in individuals with cerebral palsy. Dev Med Child Neurol 2020;62:132-9. [Crossref] [PubMed]
10. Geytenbeek J, Harlaar L, Stam M, Ket H, Becher JG, Oostrom K, Vermeulen J. Utility of language comprehension tests for unintelligible or non-speaking children with cerebral palsy: a systematic review. Dev Med Child Neurol 2010;52:1098.
11. Jacobs NPT, Pouwels PJW, van der Krogt MM, Meyns P, Zhu K, Nelissen L, Schoonmade LJ, Buizer AI, van de Pol LA. Brain structural and functional connectivity and network organization in cerebral palsy: A scoping review. Dev Med Child Neurol 2023;65:1157-73. [Crossref] [PubMed]
12. Choi JY, Choi YS, Park ES. Language Development and Brain Magnetic Resonance Imaging Characteristics in Preschool Children With Cerebral Palsy. J Speech Lang Hear Res 2017;60:1330-8. [Crossref] [PubMed]
13. Sperber C, Griffis J, Kasties V. Indirect structural disconnection-symptom mapping. Brain Struct Funct 2022;227:3129-44. [Crossref] [PubMed]
14. Na Y, Jung J, Tench CR, Auer DP, Pyun SB. Language systems from lesion-symptom mapping in aphasia: A meta-analysis of voxel-based lesion mapping studies. Neuroimage Clin 2022;35:103038. [Crossref] [PubMed]
15. Medaglia JD, Erickson BA, Pustina D, Kelkar AS, DeMarco AT, Dickens JV, Turkeltaub PE. Simulated Attack Reveals How Lesions Affect Network Properties in Poststroke Aphasia. J Neurosci 2022;42:4913-26. [Crossref] [PubMed]
16. Pretzel P, Dhollander T, Chabrier S, Al-Harrach M, Hertz-Pannier L, Dinomais M, Groeschel S. Structural brain connectivity in children after neonatal stroke: A whole-brain fixel-based analysis. Neuroimage Clin 2022;34:103035. [Crossref] [PubMed]
17. Long HL, Christensen L, Hayes S, Hustad KC. Vocal Characteristics of Infants at Risk for Speech Motor Involvement: A Scoping Review. J Speech Lang Hear Res 2023;66:4432-60. [Crossref] [PubMed]
18. Varoquaux G, Raamana PR, Engemann DA, Hoyos-Idrobo A, Schwartz Y, Thirion B. Assessing and tuning brain decoders: Cross-validation, caveats, and guidelines. Neuroimage 2017;145:166-79. [Crossref] [PubMed]
19. Hu J, Zhang J, Yang Y, Liang T, Huang T, He C, Wang F, Liu H, Zhang T. Prediction of Communication Impairment in Children With Bilateral Cerebral Palsy Using Multivariate Lesion- and Connectome-Based Approaches: Protocol for a Multicenter Prospective Cohort Study. Front Hum Neurosci 2022;16:788037. [Crossref] [PubMed]
20. Li X, Yang J, Gao J, Luo X, Zhou Z, Hu Y, Wu EX, Wan M. A robust post-processing workflow for datasets with motion artifacts in diffusion kurtosis imaging. PLoS One 2014;9:e94592. [Crossref] [PubMed]
21. Tournier JD, Calamante F, Connelly A. Robust determination of the fibre orientation distribution in diffusion MRI: non-negativity constrained super-resolved spherical deconvolution. Neuroimage 2007;35:1459-72. [Crossref] [PubMed]
22. Smith RE, Tournier JD, Calamante F, Connelly A. SIFT: Spherical-deconvolution informed filtering of tractograms. Neuroimage 2013;67:298-312. [Crossref] [PubMed]
23. Tournier JD, Smith R, Raffelt D, Tabbara R, Dhollander T, Pietsch M, Christiaens D, Jeurissen B, Yeh CH, Connelly A. MRtrix3: A fast, flexible and open software framework for medical image processing and visualisation. Neuroimage 2019;202:116137. [Crossref] [PubMed]
24. Shi Z, Sharif N, Luo K, Tan S. Development of a New Scoring System in Higher Animals for Testing Cognitive Function in the Newborn Period: Effect of Prenatal Hypoxia-Ischemia. Dev Neurosci 2025;47:12-26. [Crossref] [PubMed]
25. Perrone S, Beretta V, Tataranno ML, Tan S, Shi Z, Scarpa E, Dell'Orto V, Ravenda S, Petrolini C, Brambilla MM, Palanza P, Gitto E, Nonnis-Marzano F. Olfactory testing in infants with perinatal asphyxia: Enhancing encephalopathy risk stratification for future health outcomes. Neurosci Biobehav Rev 2025;169:106029. [Crossref] [PubMed]
26. Davis T, LaRocque KF, Mumford JA, Norman KA, Wagner AD, Poldrack RA. What do differences between multi-voxel and univariate analysis mean? How subject-, voxel-, and trial-level variance impact fMRI analysis. Neuroimage 2014;97:271-83. [Crossref] [PubMed]
27. Niddam DM, Lai KL, Hsiao YT, Wang YF, Wang SJ. Grey matter structure within the visual networks in migraine with aura: multivariate and univariate analyses. Cephalalgia 2024;44:3331024231222637. [Crossref] [PubMed]
28. Becker M, Sommer T, Kühn S. Inferior frontal gyrus involvement during search and solution in verbal creative problem solving: A parametric fMRI study. Neuroimage 2020;206:116294. [Crossref] [PubMed]
29. Zhang JX, Zhuang J, Ma L, Yu W, Peng D, Ding G, Zhang Z, Weng X. Semantic processing of Chinese in left inferior prefrontal cortex studied with reversible words. Neuroimage 2004;23:975-82. [Crossref] [PubMed]
30. Lu J, Zhao Z, Zhang J, Wu B, Zhu Y, Chang EF, Wu J, Duffau H, Berger MS. Functional maps of direct electrical stimulation-induced speech arrest and anomia: a multicentre retrospective study. Brain 2021;144:2541-53. [Crossref] [PubMed]
31. Perani D, Saccuman MC, Scifo P, Anwander A, Spada D, Baldoli C, Poloniato A, Lohmann G, Friederici AD. Neural language networks at birth. Proc Natl Acad Sci U S A 2011;108:16056-61. [Crossref] [PubMed]
32. Cona G, Semenza C. Supplementary motor area as key structure for domain-general sequence processing: A unified account. Neurosci Biobehav Rev 2017;72:28-42. [Crossref] [PubMed]
33. Glikmann-Johnston Y, Oren N, Hendler T, Shapira-Lichter I. Distinct functional connectivity of the hippocampus during semantic and phonemic fluency. Neuropsychologia 2015;69:39-49. [Crossref] [PubMed]
34. Laporta-Hoyos O, Fiori S, Pannek K, Ballester-Plané J, Leiva D, Reid LB, Pagnozzi AM, Vázquez É, Delgado I, Macaya A, Pueyo R, Boyd RN. Brain lesion scores obtained using a simple semi-quantitative scale from MR imaging are associated with motor function, communication and cognition in dyskinetic cerebral palsy. Neuroimage Clin 2018;19:892-900. [Crossref] [PubMed]
35. Sato W, Uono S. The atypical social brain network in autism: advances in structural and functional MRI studies. Curr Opin Neurol 2019;32:617-21. [Crossref] [PubMed]
36. Cao P, Gong S, Liu L, Liang G. Network-guided neuromodulation for epilepsy: Unveiling the pathway to personalized therapy. J Transl Int Med 2023;11:203-5. [Crossref] [PubMed]
37. Valeriani D, Simonyan K. The dynamic connectome of speech control. Philos Trans R Soc Lond B Biol Sci 2021;376:20200256. [Crossref] [PubMed]
38. Deco G, Tononi G, Boly M, Kringelbach ML. Rethinking segregation and integration: contributions of whole-brain modelling. Nat Rev Neurosci 2015;16:430-9. [Crossref] [PubMed]
39. Fuertinger S, Horwitz B, Simonyan K. The Functional Connectome of Speech Control. PLoS Biol 2015;13:e1002209. [Crossref] [PubMed]
40. Guo M, Cao Q, Xia S, Cao X, Chen J, Qian Y, Bao X, Xu Y. A newly-synthesized compound CP-07 alleviates microglia-mediated neuroinflammation and ischemic brain injury via inhibiting STAT3 phosphorylation. J Transl Int Med 2023;11:156-68. [Crossref] [PubMed]