Zhengwang Wu’s research while affiliated with University of North Carolina at Chapel Hill and other places

What is this page?


This page lists works of an author who doesn't have a ResearchGate profile or hasn't added the works to their profile yet. It is automatically generated from public (personal) data to further our legitimate goal of comprehensive and accurate scientific recordkeeping. If you are this author and want this page removed, please let us know.

Publications (126)


Early life phthalate exposure impacts gray matter and white matter volume in infants and young children
  • Preprint

February 2025

·

13 Reads

Emily J Werder

·

Kun Lu

·

Jake E Thistle

·

[...]

·

Stephanie Engel

Objective: Prenatal phthalate exposure is associated with adverse neurodevelopmental outcomes, yet data on impacts of early life exposure remains limited. We investigated phthalate and replacement plasticizer exposures from 2 weeks to 7 years of age in relation to brain anatomical attributes, using serial structural magnetic resonance imaging (sMRI). Material and Methods: Children were enrolled after birth into the UNC Baby Connectome Project, a longitudinal neuroimaging study. Urine samples (n=406) were collected at each visit and analyzed for 17 phthalate and replacement plasticizer metabolites. Among 157 children contributing 369 sMRIs, we calculated metabolite-specific average exposures across each individual's urine samples and used linear mixed models to estimate longitudinal associations of log transformed, specific gravity-adjusted average metabolite concentrations with gray (GMV) and white matter (WMV), and cortical volume (CV), thickness (CT), and surface area (CSA). We examined sex-specific differences in these associations. Results: Higher average metabolite concentration was associated with lower GMV (MCPP: (-1.73 cm3, 95% CI: -3.36, -0.10) and higher WMV (ΣDEHP: 2.28 cm3, 95% CI: 0.08, 4.48). Among boys (n=72, 140 sMRIs), MEP (-2.97 cm3, 95% CI: -5.85, -0.09) and MiBP (-2.40 cm3, 95% CI: -4.64, -0.15) were also associated with lower GMV. Among girls (n=85, 229 MRIs), higher ΣDINCH exposure was associated with higher WMV (2.27 cm3, 95% CI: 0.29, 4.25). We observed significant sex interactions for ΣDEHP with GMV (p-interaction=0.03) and ΣDINCH with WMV (p-interaction=0.001). Conclusion: Early life phthalate/plasticizer exposure may differentially impact various brain region volumes in early childhood, with potential downstream consequences on functional development.


Role of Data-driven Regional Growth Model in Shaping Brain Folding Patterns
  • Article
  • Full-text available

January 2025

·

29 Reads

Soft Matter

The surface morphology of the developing mammalian brain is crucial for understanding brain function and dysfunction. Computational modeling offers valuable insights into the underlying mechanisms for early brain folding. Recent...

Download

Figure 1. Dataset description. (a) Dataset construction. To study the LGM-based fingerprinting using different distributions of age and session gaps, Datasets I, II, and III consist of the first two scans, the last two scans, and the first and last scans for each subject, respectively. (b) The detailed age distribution information of the three datasets, the box plots of the scan age of session 1, session 2, and the session gap between session 1 and session 2. (c)-(e) The scan distribution of Datasets I-III, where the blue and red points represent the scans acquired from session 1 and session 2, respectively.
Figure 3. Factors affecting identification accuracy. (a) Identification using different patch sizes and patch numbers.
Figure 4. Vertex-wise contributions to identification. (a) Network-wise average uniqueness map. The color of the bars, yellow/green in the top row and black/magenta in the bottom row, indicates left and right hemispheres, respectively. (b) Network-wise average differential power map. The bars represent the mean and std. of the corresponding values across the dataset I, II, and III.
Local Gradients of Functional Connectivity Enable Precise Fingerprinting of Infant Brains During Dynamic Development

December 2024

·

17 Reads

Brain functional connectivity patterns exhibit distinctive, individualized characteristics capable of distinguishing one individual from others, like fingerprint. Accurate and reliable depiction of individualized functional connectivity patterns during infancy is crucial for advancing our understanding of individual uniqueness and variability of the intrinsic functional architecture during dynamic early brain development, as well as its role in neurodevelopmental disorders. However, the highly dynamic and rapidly developing nature of the infant brain presents significant challenges in capturing robust and stable functional fingerprint, resulting in low accuracy in individual identification over ages during infancy using functional connectivity. Conventional methods rely on brain parcellations for computing inter-regional functional connections, which are sensitive to the chosen parcellation scheme and completely ignore important fine-grained, spatially detailed patterns in functional connectivity that encodes developmentally-invariant, subject-specific features critical for functional fingerprinting. To solve these issues, for the first time, we propose a novel method to leverage the high-resolution, vertex-level local gradient map of functional connectivity from resting-state functional MRI, which captures sharp changes and subject-specific rich information of functional connectivity patterns, to explore infant functional fingerprint. Leveraging a longitudinal dataset comprising 591 high-resolution resting-state functional MRI scans from 103 infants, our method demonstrates superior performance in infant individual identification across ages. Our method has unprecedentedly achieved 99% individual identification rates across three age-varied sub-datasets, with consistent and robust identification rates across different phase encoding directions, significantly outperforming atlas-based approaches with only around 70% accuracy. Further vertex-wise uniqueness and differential power analyses highlighted the discriminative identifiability of higher-order functional networks. Additionally, the local gradient-based functional fingerprints demonstrated reliable predictive capabilities for cognitive performance during infancy. These findings suggest the existence of unique individualized functional fingerprints during infancy and underscore the potential of local gradients of functional connectivity in capturing neurobiologically meaningful and fine-grained features of individualized characteristics for advancing normal and abnormal early brain development.








Schematic diagram of this research. (a) Cortical heterogeneous growth map with 18 regions determined based on the developmental patterns of surface area from 29 post-menstrual weeks to 24 postnatal months of age; (b). Symbolic regression algorithms used to identify the mathematical growth models for each region, exemplified with region 4; (c). Construction process of finite element model for each region, ROI means region of interest. (d). Qualitative and quantitative comparison between the simulation results and real human brains measures along the developmental timeline.
Role of Data-driven Regional Growth Model in Shaping Brain Folding Patterns

August 2024

·

42 Reads

The surface morphology of the developing mammalian brain is crucial for understanding brain function and dysfunction. Computational modeling offers valuable insights into the underlying mechanisms for early brain folding. While previous studies generally assume uniform growth, recent findings indicate significant regional variations in brain tissue growth. However, the role of these variations in cortical development remains unclear. In this study, we explored how regional cortical growth affects brain folding patterns. We first developed growth models for typical cortical regions using ML-assisted symbolic regression, based on longitudinal data from over 1,000 infant MRI scans that captured cortical surface area and thickness during perinatal and postnatal brains development. These models were subsequently integrated into computational software to simulate cortical development with anatomically realistic geometric models. We quantified the resulting folding patterns using metrics such as mean curvature, sulcal depth, and gyrification index. Our results demonstrate that regional growth models generate complex brain folding patterns that more closely match actual brains structures, both quantitatively and qualitatively, compared to uniform growth models. Growth magnitude plays a dominant role in shaping folding patterns, while growth trajectory has a minor influence. Moreover, multi-region models better capture the intricacies of brain folding than single-region models. Our results underscore the necessity and importance of incorporating regional growth heterogeneity into brain folding simulations, which could enhance early diagnosis and treatment of cortical malformations and neurodevelopmental disorders such as epilepsy and autism.


Citations (60)


... Temporal lobe. The temporal lobe is the part of the brain responsible for various cognitive functions including memory, senses, auditory, language processing, cognition, and semantics (19). Some studies report that while the temporal lobe has a larger area especially on the left side, on the contrary, the cortex thickness is higher on the right side (20,21). ...

Reference:

Glioma lateralization: Focus on the anatomical localization and the distribution of molecular alterations (Review)
Longitudinally Consistent Registration and Parcellation of Cortical Surfaces Using Semi-Supervised Learning
  • Citing Article
  • May 2024

Medical Image Analysis

... Path signatures (Chevyrev and Kormilitzin, 2016), as effective descriptors of ordered data, capture essential characteristics of trajectories and have been successfully applied in various domains of neuroscience. For instance, path signatures have been employed to predict Alzheimer's diagnosis by modeling disease progression trajectories (Moore et al., 2019), to detect epileptic seizures by analyzing electroencephalogram (EEG) patterns (Tang et al., 2024), in early autism diagnosis through behavioral pattern recognition (Yin et al., 2024), and in seizure forecasting (Haderlein et al., 2023). ...

Early autism diagnosis based on path signature and Siamese unsupervised feature compressor
  • Citing Article
  • May 2024

Cerebral Cortex

... Figure 3 shows the structure of a U-Net network. Scholars have proposed many improved and derived segmentation models based on the U-Net model, such as those of Payette et al. [38], Khalili et al. [32], Zhao et al. [39], Peng et al. [35] and Pei et al. [40]. Payette et al. focused on the problem of noise labels, used noise labels to train the U-Net network, and introduced transfer learning to further improve the segmentation results. ...

PETS-Nets: Joint Pose Estimation and Tissue Segmentation of Fetal Brains Using Anatomy-Guided Networks
  • Citing Article
  • October 2023

IEEE Transactions on Medical Imaging

... This model incorporates an Orthonormal Clustering Readout operation, resulting in cluster-aware node embeddings and informative graph embeddings. Fang et al. [27] introduced the Path-based Heterogeneous Brain Transformer Network (PH-BTN), which constructs brain graphs from rs-fMRI data and learns compact edge features through heterogeneous graph paths, enhancing brain network analysis. Other approaches address specific challenges in brain network analysis. ...

Path-Based Heterogeneous Brain Transformer Network for Resting-State Functional Connectivity Analysis
  • Citing Chapter
  • October 2023

Lecture Notes in Computer Science

... Similar findings are also reported for MSEL-based future (at 4 years of age) cognitive score prediction using sMRI brain features at birth, such as cortical thickness, mean curvature, local gyrification index, vertex area, vertex volume, sulcal depth in string distance and sulcal depth in Euclidean distance with a mean root square error of 0.023-0.18 (Adeli et al., 2019;Zhang et al., 2018Zhang et al., , 2020Cheng et al., 2022Cheng et al., , 2023. Furthermore, Wang et al. (2015) used multi-kernel SVR for estimating IQ values using cortical thickness, surface area, sulcal depth, and curvature from BAs 1, 2, 3, 4, 7, 32, 34, 39, 40, 41, 42, 44, 45, and 47 and obtain an average correlation coefficient of 0.718 and a mean root mean square error of 8.695 between the true FSIQs and the estimated ones. ...

Prediction of Infant Cognitive Development with Cortical Surface-Based Multimodal Learning
  • Citing Chapter
  • October 2023

Lecture Notes in Computer Science

... This method was evaluated on healthy controls and multiple sclerosis (MS) cohorts. Zhao et al. developed a deep learning model to harmonize the multi-site cortical data using a surface-based autoencoder [104]. The encoded cortical features were subsequently decomposed into components related to site-specific characteristics and those unrelated to site effects. ...

Disentangling Site Effects with Cycle-Consistent Adversarial Autoencoder for Multi-site Cortical Data Harmonization
  • Citing Chapter
  • October 2023

Lecture Notes in Computer Science

... NeSVoR [18] was developed as a novel slice-to-volume reconstruction method using implicit neural representation. Ma et al. [19] proposed a deep learning model for motion correction based on the relative motion position between adjacent slices. These above works all use deep learning-based methods to predict camera or slice localization/position in 3D space based on image intensity information. ...

Geometric Constrained Deep Learning for Motion Correction of Fetal Brain Mr Images
  • Citing Conference Paper
  • April 2023

... The previous studies on the ALFF and fALFF values in POAG patients were mostly based on volumetric spatial calculations. Previous studies demonstrate that, in capturing and preserving the precise boundaries of cortical regions with volume-based processing methods, surface-based approaches prove to be more effective, enhancing spatial precision in the representation of brain structures (Wang, 2023). The effectiveness of brain surface analysis methods is three times higher than that of traditional volumetric methods (Wang, 2023, Esteban et al., 2019. ...

Fine-grained functional parcellation maps of the infant cerebral cortex
  • Citing Article
  • August 2023

... Traditionally, FC fingerprinting analyzes brain connectivity by mapping inter-region connections between a set of region-of-interests (ROIs) from predefined brain atlases, such as those containing 268 [1] , 333 [13] , 602 ROIs [14] or over 1000 ROIs [15,16], which have been widely applied to reveal complex functional patterns in both mature and immature brains. In addition to atlas-based methods, numerous studies have employed principal gradient techniques, a vertex-wise approach, to capture global, brain-wide patterns and macroscale connectivity transitions along the cortical surface [17], as network topology has demonstrated significant behavioral implications [13,18,19]. ...

Fine-grained functional parcellation maps of the infant cerebral cortex
  • Citing Article
  • August 2023

... Another source of variance could be small, but real, brain developmental changes across the recording days. Given the reliability limitations in infant precision functional imaging, working at the level of infant specific parcels (Myers et al., 2024;Wang et al., 2023) or individual networks , could help to improve reliability at the expense of spatial precision. Another approach would be to use alternative methods to increase overall SNR, such as higher magnetic fields (Annink et al., 2020), which could mitigate partial volume effects by allowing for smaller voxel sizes more fitting to the smaller infant brains, or infant specific coils (Hughes et al., 2017;Keil et al., 2011;Lopez Rios et al., 2018). ...

Fine-grained functional parcellation maps of the infant cerebral cortex

eLife