Assessing the impact of fetal growth restriction using high-resolution MRI: a comprehensive cortical analysis

Abstract

Background

Fetal Growth Restriction (FGR) is a complex neurodevelopmental condition marked by the failure of a fetus to reach its genetically determined growth potential, leading to significant neurodevelopmental risks. Emerging MR scans exhibit superiority to traditional US-based diagnostic approaches, while quantitative morphological analysis of FGR is still lacking.

Methods

This study explores the effects of FGR on fetal brain development using advanced Magnetic Resonance Imaging (MRI) techniques. A comprehensive analysis of 35 brain regions was conducted, focusing on detailed cortical properties such as curvature, sulcal depth, surface area, cortical gray matter volume, and cortical thickness, while distinguishing between the left and right hemispheres. Our comprehensive analysis leverages super-resolution reconstruction, segmentation, and surface reconstruction algorithms to obtain statistical data.

Results

Our analysis revealed that (1) FGR equally affected both brain hemispheres; (2) In the absence of differentiation between brain regions, different cortical metrics had no significant effect on FGR; and (3) the most significant cortical metrics were primarily observed in cortical thickness and sulcal depth; (4) a regression model based on only seven key brain regions could effectively predict FGR. The findings demonstrate the value of high-resolution MRI in detecting early biomarkers for FGR, paving the way for improved prenatal care and targeted therapeutic strategies.

Conclusions

The findings highlight the importance of high-resolution MRI in identifying early biomarkers for FGR. The study emphasizes the value of advanced cortical analysis in understanding the pathophysiological mechanisms underlying FGR and offers a robust framework for early diagnosis and intervention. These insights could contribute to improved prenatal care and targeted therapeutic strategies, ultimately enhancing outcomes for affected populations.

Introduction

Fetal Growth Restriction (FGR) is a complex, pathologic condition characterized by the failure of a fetus to achieve its genetically predetermined growth potential [1]. According to a large-scale study by Andreasen et al. (2021), among 78,544 pregnancies, 3,069 (3.9%) were diagnosed with fetal growth restriction [2]. Globally, FGR impacts approximately 30 million infants annually [3]. FGR has significant and long-lasting effects on the structure and function of developing brain [4]. Infants with FGR often exhibit reduced overall brain volume [5], altered cortical structure, and impaired brain connectivity. These structural anomalies manifest in neuronal migration deficits, reduced dendritic processes, and decreased efficiency in long-range neural networks, particularly in the prefrontal and limbic regions [6]. The long-term functional effects of these changes are evident in school-aged children, who frequently exhibit deficits in cognition, fine motor skills, memory, and neuropsychological functions, accounting for the poor attention, hyperactivity, and altered mood [7–11]. Moreover, FGR is associated with a heightened risk of preterm birth, perinatal death, and neurodevelopmental deficits. Studies indicate that both early-onset (< 32 GW) and late-onset (> 32 GW) FGR are the significant risk factors for cerebral palsy and other neurocognitive impairments [3, 12]. The hemodynamic changes in cerebral perfusion associated with FGR further exacerbate these risks, as chronic hypoxia and the resulting cardiovascular redistribution can lead to asymmetric fetal growth and long-lasting neurological damage. The causes of FGR are diverse, encompassing fetal, maternal, and placental factors [2, 13]. In summary, given the significant risks posed by FGR, especially its impact on neurodevelopment, further research is crucial to better understand its mechanisms and develop more effective intervention strategies.

Existing diagnostic approaches, predominantly ultrasound [14, 15], provide critical insights into biometric parameters, yet often lack the resolution necessary for a detailed investigation into the underlying mechanisms of fetal brain development, particularly in conditions like FGR. Recent advancements in Magnetic Resonance Imaging (MRI) have not only improved the accuracy of prenatal diagnostics but also offered unprecedented insights into the pathophysiological processes affecting fetal brain development [16–18]. While MRI-related FGR research has primarily focused on observing changes in total brain volume [4, 19–21], there has been limited exploration into the detailed cortical properties affected by FGR. In this study, we extend the application of MRI beyond the traditional emphasis on regional brain volume measurements to examine advanced cortical properties such as curvature, sulcal depth, surface area, gray matter volume, and cortical thickness. By investigating these features, we aim to reveal the underlying mechanisms of cortical development and surface growth patterns in fetuses affected by FGR, providing a more comprehensive understanding of how this condition impacts brain structure.

Materials and methods

This retrospective study was conducted after obtaining approval from the Institutional Review Board of our medical center. We systematically retrieved and analyzed MR images from the Picture Archiving and Communication System (PACS) of singleton pregnant women who underwent MR scanning for fetal evaluation at the First Affiliated Hospital of Nanjing Medical University, from June 2018 to May 2023 (Fig. 1). To ensure data quality, we carefully selected high-quality scans and applied advanced reconstruction algorithms, as outlined in PAK-SRR, to mitigate issues with low-quality images. Scans with reconstruction failures were manually excluded, ensuring that only reliable data were included in our analysis. These steps were taken to minimize the impact of poor image quality and reconstruction errors on the reliability of our results.

Fig. 1.

Data inclusion and age distribution. a Flowchart of data inclusion and exclusion; b The frequency distribution of GA of fetuses was finally included in this study

Subjects

Inclusion Criteria for FGR Subjects $\to$ 83 subjects: The study included singleton pregnancies, for which fetal MR scans are available for analysis. The criteria for inclusion in the FGR [17] group are as follows:

• MRI scans were conducted when the GA was between 27 and 39 weeks.

• FGR pre-diagnosed via ultrasound, with neonatal birth weight below the 10 th percentile.

• Evidence of placental insufficiency, characterized by one or more of the following concerns: high umbilical artery pulsatility index, elevated umbilical artery arterial resistance, abnormal systolic-diastolic ratio (exceeding 3% or showing absence of end-diastolic flow or reverse flow), increased diastolic blood pressure in the umbilical artery, and abnormal middle cerebral artery (MCA) pulsatility index or ultrasound indications of placental lesions, thickening, or low amniotic fluid index.

Inclusion Criteria for Healthy Controls (HCs) $\to$ 158 subjects: The HCs group comprised pregnant women whose fetuses showed no brain pathologies on MR scans and who had normal biometric brain values that matched GA, as determined from biometric population studies. HCs were selected to provide a baseline for normal fetal brain development, facilitating comparative analyses with the FGR group.

Concurrently, other potential FGR etiologies, such as intrauterine infections, genetic factors, and multifetal pregnancies, have proved inapplicable. MRI is the preferred investigative modality for further investigation in patients with FGR after abnormal findings on ultrasound. HCs group patients underwent MRI due to histories of abnormal pregnancies, anomalies detected in prior examinations, or atypical prenatal neurological assessments. Pediatric radiologists discerned no pathologic abnormalities in the MRI evaluation.

Exclusion Criteria for FGR and HCs Subjects $\to$ 11 FGR and 15 HCs subjects: Subjects are excluded based on the following criteria:

• Fetal causes, including the presence of intrauterine infections, abnormal fetal anatomical findings, abnormal genetic test results, or multiple gestations.

• Maternal causes, such as uterine malformations, chronic cardiovascular diseases, other significant maternal illnesses, or chronic alcohol and drug abuse.

• Insufficient clinical information or data related to the pregnancy.

• Poor quality of MR scans that precluded reliable analysis.

• Failure in super-resolution or surface reconstruction.

Measurements

Ethics

The study was approved by the local institution, and the registration number was 2021-SR-407. The medical information collected will be kept confidential and not passed on to anyone who does not belong to the study. The information is displayed anonymously without revealing the identities of the study participants.

MR imaging protocol

MRI was performed on a 1.5 T MRI system (MAGNETOM Aera, Siemens Healthineers AG, Erlangen, Germany) equipped with a 6-channel body coil. Neither the mother nor the fetus received sedation, and the mother maintained free-breathing throughout scanning. The MRI protocol included (1) axial, coronal, and sagittal T2-weighted rapid acquisition HASTE (Half Fourier Acquisition Single Shot Turbo Spin Echo) sequence with the following parameters: the size of field of view (FOV): $320\times 320$ mm, 3 mm thick slice, matrix size is $256\times 256$; time of repetition (TR): 1200 ms, time of echo (TE): 168 ms. (2) Axial T1 WI vibe with the following parameters: the FOV size $380\times 369$ mm, 3 mm thick slice, matrix size $320\times 240$, TR: 6.83 ms; TE: 2.39 ms and 4.77 ms. (3) A DWI sequence with two different b-values (50 and 800 s/${\mathrm{mm}}^2$). Twenty-five 3 mm thick slices are collected; the matrix size is $114\times 105$; the FOV size is $320\times 320$ mm. T2-weighted sequences are used for image reconstruction, and other sequences exclude cases with lesions such as brain malformations, hemorrhages, or tumors.

Data preprocessing

Due to irregular fetal movement, clinically acquired fetal brain MR images exhibited multiple orientations with varying resolutions (0.8$\sim$1.2 mm) and thick-slice scans (thickness > 3 mm) [22]. To address this challenge, we employed the PAK-SRR [23] and NeSVoR [24] techniques to reconstruct high-resolution MR images (with a spatial resolution of $0.8\times 0.8\times 0.8$ mm) from clinically acquired multi-view thick-slice scans. Clinicians then selected the highest-quality reconstructed images from the results of these two methods. We then segmented white matter (WM), gray matter (GM), and cerebrospinal fluid (CSF) using HieraParceNet [25]. HieraParceNet is a state-of-the-art method specifically developed for fetal brain segmentation, capable of handling brain images from fetuses at different gestational ages. The pre-segmented images were subsequently processed with Freesurfer [26] to reconstruct the brain cortical surface and further segment into 35 left brain regions and 35 right brain regions [27], as demonstrated in Fig. 2. Finally, we calculated the cortical properties from the reconstructed cortical surface, including (1) gray volume, (2) sulcal depth, (3) cortical thickness, (4) curvature, and (5) surface area. The data analysis pipeline is shown in Fig. 3.

Fig. 2.

Brain parcellation of gray matter, encompassing 35 symmetric brain regions with differentiation between left and right hemispheres

Fig. 3.

Data analysis pipeline

Cortical properties

In this study, we utilized five types of cortical properties to assess fetal brain development and constraints, including:

• Gray Volumes: Gray volumes refer to the total volume of gray matter within specific brain regions, primarily composed of neuronal cell bodies, responsible for processing information and executing brain functions [28].

• Suicl Depth: Sulcal depth measures the depth of the gyrus and sulcus on the brain’s surface. These folds are vital for increasing the brain’s surface area, and enhancing its ability to manage complex cognitive tasks [29].

• Cortical Thickness: Cortical thickness refers to the thickness of the cerebral cortex, typically measured in millimeters and related to the density of neurons and the complexity of neuronal connections [30, 31].

• Cortical Curvature: Curvature describes the bending and curving of the cortical. It reflects the morphological features of the cerebral cortex, including the formation of gyrus and sulcus, which is crucial for accommodating cognitive functions and neural connectivity [32, 33].

• Surface Area: Surface area refers to the expanse of the cerebral cortex, correlating with the brain’s capacity to process information and maintain extensive neural networks. A larger surface area typically reflects greater computational ability and complex cognitive functions [34].

These cortical properties collectively reflect the structural and developmental status of the brain, providing comprehensive information for in-depth study of the effects of FGR. Through analysis of these properties, we can better understand the pathophysiological mechanisms of FGR and provide critical clinical references for early diagnosis and intervention. It is important to note that the main purpose of this study is not to explore the underlying causes of FGR but rather to detect and describe the characteristics of FGR based on high-resolution MRI data. We aimed to provide a detailed analysis of the cortical changes associated with FGR, with a particular focus on the imaging features, rather than delve into the complex factors that may contribute to the condition.

Statistics

We analyzed the global cortical properties to assess their relationship with FGR. Starting with ANOVA, we treated each group as independent variables, cortical features as dependent variables and GA as a covariate to identify statistically significant features (p value < 0.05). We then apply variable correlation screening for these significant features, retaining only one feature from each pair with a correlation coefficient greater than 0.9, selected randomly. Subsequently, we conducted binary logistic regression with gestational age (GA) as a covariate and applied False Discovery Rate (FDR) correction to calculate the odds ratios (OR) and p values, to determine whether these features protective or suppressive impacted FGR. It is important to clarify that the designations of ‘protective’ and ‘suppressive’ are based solely on statistical relationships observed in the data, comparing FGR cases to controls. While there is potential bias, we believe that our relatively large and well-characterized sample size ensures our findings to be robust and representative of typical cortical changes associated with FGR. Finally, we assessed the diagnostic accuracy of the regression model by calculating the AUC.

Our study included four comparative experiments. (1) We analyzed cortical features separately for the left and right hemispheres to determine whether their influence on FGR was symmetric. (2) We aggregated features from 5 cortical properties to examine their combined effect on FGR. (3) We expanded our feature set to 175 by combining 35 brain regions with five cortical properties. (4) Finally, we increased this set to 350 features by differentiating between the left and right hemispheres for further analysis.

Statistical analyses are performed using algorithms consistent with SPSS and reproduced using Python, facilitating streamlined analysis, and the binary logistic regression is used as the classification model. The code is openly accessible at https://github.com/sj-huang/FGR_cortex.

Results and analysis

Data description:

The statistical analysis presented in Table 1 evaluates the biometric differences between the FGR and HCs groups across various gestational weeks. Throughout the observed range, the T values predominantly indicate a negligible difference between the two groups, corroborating the closeness of their respective measurements. Notably, the T values are relatively low (ranging from −1.46 to 1.38), suggesting a substantial similarity between FGR and HCs groups. This similarity is further substantiated by the p values (> 0.05). The highest T value observed is 1.38, with a p value of 0.17, lacking significant statistical differences between the groups. This consistent pattern of non-significant differences highlights the similarity between the FGR and HC groups in biometric parameters, with no notable variation across the gestational timeline. Totally 241 fetuses are included in the study group. The mean GA in which the MR imaging is performed in the control group is 31.87 (standard deviation (SD), 3.67) weeks and the study group is 31.45 (SD, 3.79). The distribution of MR images according to GA is shown in Fig. 1.

Table 1.

The detailed comparison of GA (day) distribution and maternal age between the FGR and HCs groups. It includes the overall average and standard deviation (SD) for both groups, as well as the median value and range. The GA distribution is further divided into individual gestational weeks (GWs)

		Group
		FGR		HCs
		Average ± SD	Median (range)	Average ± SD	Median (range)	p value	T value
GA (day)		231.02 ± 20.52	228.51 (189.0 - 270.0)	234.72 ± 18.65	238.0 (192.0 - 275.0)	0.17	1.38
Maternal age (year)		31.45 ± 3.79	31.79 (22.0 - 38.0)	31.87 ± 3.67	32.0 (21.0 - 39.29)	0.40	0.83
GA (day)	27 GWs	189.00 ± NAN	189.0 (189.0 - 189.0)	192.00 ± NAN	192.0 (192.0 - 192.0)	NAN	NAN
	28 GWs	196.66 ± 0.57	196.98 (196.0 - 197.0)	196.34 ± 1.37	196.0 (195.0 - 199.01)	0.63	−0.50
	29	202.20 ± 1.30	203.0 (200.0 - 203.0)	202.39 ± 1.68	202.0 (200.97 - 205.0)	0.85	0.20
	30 GWs	210.16 ± 1.40	210.0 (207.0 - 212.0)	210.75 ± 1.42	210.99 (208.0 - 213.0)	0.32	1.02
	31 GWs	217.28 ± 1.90	217.0 (214.97 - 220.0)	217.50 ± 1.90	217.0 (213.99 - 220.01)	0.81	0.25
	32 GWs	224.64 ± 1.60	224.0 (222.0 - 227.0)	223.94 ± 1.78	224.0 (220.99 - 226.94)	0.25	−1.17
	33 GWs	231.17 ± 0.98	231.0 (230.02 - 233.0)	231.31 ± 1.45	231.0 (228.0 - 234.0)	0.80	0.26
	34 GWs	238.50 ± 1.72	238.5 (235.0 - 241.01)	238.53 ± 2.07	238.0 (235.0 - 241.01)	0.97	0.04
	35 GWs	244.50 ± 1.38	245.0 (242.0 - 246.0)	245.24 ± 1.23	245.0 (243.0 - 248.01)	0.27	1.21
	36 GWs	252.43 ± 1.13	252.0 (252.0 - 255.0)	251.59 ± 1.79	252.0 (248.99 - 255.01)	0.16	−1.46
	37 GWs	259.56 ± 1.51	259.0 (257.0 - 262.0)	258.77 ± 2.21	259.0 (255.99 - 262.01)	0.32	−1.00
	38 GWs	264.75 ± 1.50	265.0 (263.0 - 266.0)	265.50 ± 2.59	265.0 (263.0 - 269.0)	0.58	0.58
	39 GWs	270.00 ± NAN	270.0 (270.0 - 270.0)	275.00 ± NAN	275.0 (275.0 - 275.0)	NAN	NAN

Symmetry analysis

In this section, we examined whether the above-mentioned five cortical features of the left and right hemispheres exhibited symmetrical effects in FGR, as shown in Table. 2. The results are presented in the table below, summarizing the p values, ORs, and the area under the ROC curve for both hemispheres.

Table 2.

Table showing Hemisphere, p value, OR, and ROC AUC

Hemisphere	p value	OR	95% CI	AUC
Right	1.23e-33	0.995	0.994 - 0.996	0.510
Left	7.54e-24	1.006	1.005 - 1.007

Both hemispheres demonstrated extremely low p values, indicating that their cortical features are significantly associated with FGR. However, the odds ratios (OR = 0.994 for the right hemisphere and OR = 1.005 for the left hemisphere) suggest that the influence of these features on FGR is nearly neutral, with slight variations around 1.0. Moreover, the AUC of the regression model is 0.51, indicating that hemispheric features alone are insufficient to distinguish between the FGR and HCs groups.

These findings confirm that the impact of hemispheric cortical features on FGR is symmetric, allowing us to combine the features of the left and right hemispheres in subsequent analyses without introducing bias.

Cortical properties analysis

In this section, we examined the impact of 5 cortical properties on FGR, as shown in Table 3. For this experiment, we observe three features (i.e., Surface Area, Curvature, and Gray Volume) that demonstrated a statistically significant effect (P < 0.05), as detailed in the table below.

Table 3.

Table showing Property, p value, OR, and ROC AUC

Property	p value	OR	95% CI	AUC
Surface Area	8.20e-5	1.002	1.001 - 1.003	0.650
Curvature	2.13e-17	1.001	1.001 - 1.001
Gray Volume	2.45e-32	0.992	0.991 - 0.994

Similar to the symmetric analysis, the selected properties exhibit very low p values, yet the ORs indicate that the influence of these features on FGR is nearly neutral. This suggests that when these cortical properties are analyzed collectively, distinguishing specific brain regions, may be insufficient to identify significant risk factors or distinguish between protective and inhibitory effects. The regression model incorporating these features yielded an AUC of 0.651, indicating a moderate ability to distinguish between the FGR and HCs groups based on these cortical properties. To better understand the relationship between cortical characteristics and FGR, it is essential to consider detailed brain regions, which may better reveal the subtle differences in how these characteristics influence FGR and uncover more diverse patterns of positive and negative effects.

Influence of combined brain regions and cortical properties on FGR

In this section, we investigated the influence of 175 features derived from combining 35 brain regions with 5 cortical properties on FGR, ultimately identifying 62 key features by performing the method described in “Statistics” section, the results shown in Fig. 4. The analysis revealed various statistically significant associations between these features and FGR.

Fig. 4.

Illustration of log(OR) and p value of the 62 selected key features (with p value < 0.05) across 5 cortical properties (i.e., Gray Volume, Curvature, Sulcal Depth, Surface Area, and Cortical Thickness), alongside the correlation coefficient matrix of all 175 features. The 62 key features were selected from the 175 features, ensuring a correlation coefficient threshold of less than 0.9

Among the most significant findings, the limbic lobe’s cortical thickness and the parietal lobe’s sulcal depth demonstrated notable protective effects. In contrast, the cortical thickness of the occipital and temporal lobes, along with the sulcal depth of the limbic lobe, exhibited apparent inhibitory effects. Conversely, measures such as gray volume, curvature, and surface area showed only weak protective or inhibitory effects, as indicated by ORs around 1. Specifically, the ORs for curvature and surface area were predominantly above 1, suggesting slight inhibitory effects. Only one brain region showed a protective effect in gray volume, with an OR of 0.986. The diagnostic performance of the regression model was robust, achieving an AUC of 0.880, indicating strong predictive accuracy across these 62 key features. This high level of accuracy reflects the model’s strong discriminatory power in differentiating between FGR and HCs, providing a reliable tool for early detection and intervention.

In “Region-specific impact with high and low ORs” section, we primarily analyze the features with very high and very low OR values from this section, and find that these features exhibit strong protective or inhibitory effects in FGR. These features are mainly concentrated in the thickness of the cuneus (OR = 53.53), temporal pole (OR = 37.14), rostral anterior cingulate (OR = 0.12), and parahippocampal (OR = 0.01). Additionally, high OR values are observed in the sulcal depth of the insula (OR = 24.64) and rostral anterior cingulate (OR = 25.08), as illustrated in Fig. 4.

Influence of distinct hemispheric brain regions features on FGR

In this section, we conducted our analysis by distinguishing specific cortical features between the left and right hemispheres, and examining the influence of 350 features on FGR. Among the initial 350 features, the majority showed strong correlations. After extensive calculations and selection processes (as described in “Statistics” section), we ultimately identified 7 essential features, as shown in Table. 4 and Fig. 5. These features include areas within the limbic, parietal, and occipital lobes across both hemispheres, primarily focusing on the sulcal depth near the central sulcus and lateral fissure, as depicted in Fig. 6. Notably, the cortical thickness of the right parahippocampal gyrus demonstrated a pronounced inhibitory effect, with an OR of 53.697. In contrast, the sulcal depth of the left insula exhibited a robust protective effect, with an OR as low as 0.092. The other features also showed significant protective and inhibitory impacts.

Table 4.

Quantitative evaluation of seven critical cortical features from distinct hemispheric brain regions associated with FGR, including regions across limbic, parietal, and occipital lobes. The table shows the anatomical region, lobe, cortical property, hemisphere, statistical significance (p value), odds ratio (OR), and 95% confidence interval (CI) for each feature. Collectively, these features achieve an AUC of 0.834, demonstrating high model efficacy in identifying FGR-related cortical features

Brain region	Lobe	Property	Hemisphere	p value	OR	95% CI	AUC
insula	Limbic	Sulcal Depth	Left	0.0229	0.092	0.012 - 0.718	0.834
precuneus	Parietal	Sulcal Depth	Right	0.0421	0.154	0.025 - 0.935
postcentral	Parietal	Curvature	Left	0.0040	0.897	0.833 - 0.966
lingual	Occipital	Sulcal Depth	Right	0.0340	3.811	1.107 - 13.125
pericalcarine	Occipital	Sulcal Depth	Left	0.0074	5.614	1.590 - 19.820
supramarginal	Parietal	Sulcal Depth	Left	0.0161	20.953	1.761 - 249.378
parahippocampal	Limbic	Cortical Thickness	Right	0.0053	53.697	3.271 - 881.456

Fig. 5.

Illustration of p value and OR value for key features (with p value < 0.05) across 5 cortical properties (i.e., Gray Volume, Curvature, Sulcal Depth, Surface Area, and Cortical Thickness). The key features were selected from the 350 features, ensuring a correlation coefficient threshold of less than 0.9

Fig. 6.

Sankey diagram for key features (with p value < 0.05) across 5 cortical properties (i.e., Gray Volume, Curvature, Sulcal Depth, Surface Area, and Cortical Thickness). The key features were selected from the 350 features, ensuring a correlation coefficient threshold of less than 0.9

In conclusion, this experiment has provided more profound insights into the influence of cortical features from different hemispheres on FGR. The findings emphasize the importance of considering both hemispheres and specific brain regions when analyzing the impact of cortical properties on FGR, which could have significant implications for early detection and intervention strategies.

Discussion

In this study, we comprehensively analyzed cortical development in FGR from four perspectives. Our experiments revealed that both the left and right hemispheres are equally affected, and a detailed parcellation of brain regions is necessary to predict FGR accurately. Additionally, the research provided more profound insights into the influence of cortical properties on FGR. Previous studies [35, 36] consistently demonstrate that fetuses with FGR exhibit delayed and abnormal cerebral cortex development. Notably, infants with FGR were found to have 28% less gray matter volume than healthy infants. This reduction in gray matter volume is likely attributed to placental dysfunction, which restricts the fetus’s access to essential nutrients, thereby disrupting neuronal production and migration, ultimately leading to cortical maldevelopment. Supporting evidence [37–39] shows that FGR rats and babies have significantly reduced neuronal density and dendritic spine protrusions at birth, along with delayed cortical folding and reduced gray matter volume.

Region-specific impact with high and low ORs

(1) The high OR values for the insula and rostral anterior cingulate indicate that structural and functional abnormalities in these regions occur significantly more frequently in the FGR group compared to the normal group. Both the insula and the rostral anterior cingulate are early-developing, metabolically active brain regions that rely heavily on sufficient nutrient and oxygen supply during fetal development [40, 41]. These areas play critical roles in processing emotions, pain, autonomic regulation, and monitoring internal states. FGR triggers an excessive stress response in these regions, leading to abnormal increases in sulcal depth. Furthermore, synaptogenesis may increase under chronic stress, but synaptic pruning is impaired, resulting in deeper cortical folds. The functional complexity and high metabolic activity of the insula and rostral anterior cingulate make them particularly sensitive to adverse environments, leading to the formation of deeper sulci [42, 43]. Research [44] suggests that fetuses with FGR are more likely to develop emotional and behavioural problems after birth, potentially due to the critical role of the insula and rostral anterior cingulate in emotion regulation and sensory processing.

(2) The high OR values for the cuneus and temporal pole suggest cortical thickening in these regions in FGR fetuses, potentially due to maladaptive neurodevelopmental responses to prolonged nutrient and oxygen deprivation. Cortical thickening in FGR may result from impaired synaptic pruning, rather than early neurodevelopmental abnormalities. Synaptic pruning is a critical process for refining neural networks, and its disruption later in development could lead to cortical thickening. In contrast, early neurodevelopmental issues tend to have more widespread, lasting effects, often causing cognitive delays, motor impairments, and other neurological conditions. However, many infants with FGR can recover normal development with appropriate care. The cuneus is involved in visual processing, particularly in interpreting spatial information and integrating visual stimuli. Thickening in this region could impair visual attention, perception, and visuospatial coordination [45]. The temporal pole, on the other hand, plays a significant role in emotion regulation, memory, and social cognition. Increased cortical thickness here could disrupt the processing of emotional responses, memory recall, and social interactions, potentially leading to cognitive and behavioral issues as the child develops [46–48].

(3) The low OR values for the rostral anterior cingulate and parahippocampal suggest cortical thinning in these regions in FGR fetuses, may arise from long-term oxygen and nutrient deprivation, which disrupts neurogenesis and synaptic connectivity. This impairment can lead to reduced neuronal density and cortical structure alterations, adversely affecting the normal development of these brain regions [49]. The rostral anterior cingulate cortex is critical for emotion regulation, decision-making, and conflict monitoring. Thinning in this region may impair the processing of emotional stimuli, potentially increasing susceptibility to stress, anxiety, and depression [50]. The parahippocampal cortex plays a crucial role in memory encoding, spatial navigation, and contextual processing. Cortical thinning in this region is associated with memory impairments, particularly in forming and retrieving memories, as seen in various neurodevelopmental and neurodegenerative conditions [51, 52]. Abnormalities in these regions may result in long-term cognitive, emotional, and behavioral challenges in infants affected by FGR, as these regions are essential for integrating sensory inputs, managing emotional responses, and forming complex memories.

Key brain regions with concentrated functions and spatial distribution

The findings presented in Table 4 highlight significant associations between FGR and structural changes in specific brain regions. Notably, these regions, including the insula, precuneus, postcentral gyrus, lingual gyrus, pericalcarine, supramarginal gyrus, and parahippocampal gyrus, are spatially distributed in a manner that suggests functional and anatomical connectivity, particularly across the parietal, occipital, and limbic lobes, as shown in Fig. 6.

Interestingly, these critical brain regions exhibit robust connectivity and centralization in both their spatial distribution and functional roles. (1) The parietal regions, particularly the precuneus, postcentral gyrus, and supramarginal gyrus, are central to sensory integration, self-awareness, and language processing [53]. (2) The occipital regions, including the lingual gyrus and pericalcarine area, are critical for visual functions [54–56]. (3) The limbic structures, like the insula and parahippocampal gyrus, further highlight the potential impact of FGR on emotional regulation and memory [51, 57]. Together, these findings suggest that FGR may lead to a cascade of developmental changes across multiple cortical regions, potentially resulting in long-term cognitive, sensory, and emotional challenges.

FGR may not affect all areas of the brain uniformly, but instead exhibits regional heterogeneity [58]. For example, areas with highly metabolically active metabolism, complex developmental processes, and sensitivity to nutrient needs are more susceptible to FGR. Conversely, regions that are less involved in the stress response, have relatively low metabolism, or have a high network redundancy, may show a relatively “protective” effect.

Clinical Implications and Application in Prenatal Care

The findings of this study provide valuable clinical insights for obstetricians and midwives in managing pregnancies affected by FGR. While this research does not offer real-time diagnostic tools, it presents critical insights into cortical characteristics that can serve as the foundation for earlier diagnosis and prenatal care. By identifying key cortical biomarkers, such as cortical thickness and sulcal depth in specific brain regions, we highlight areas of the brain that may be particularly vulnerable in fetuses with FGR. These biomarkers offer clinicians a deeper understanding of fetal brain development and provide essential information for the future early identification of growth-restricted fetuses.

Our study can inspire more accurate and earlier diagnoses and prenatal surveillance strategies for FGR. For instance, if abnormal development is observed in regions associated with motor control, it may signal potential delays in developmental milestones, such as delayed motor skills after birth. Monitoring motor-related brain regions more closely during early-stage ultrasound imaging and other prenatal diagnostic procedures can improve the accuracy of early FGR diagnosis, allowing for more timely and effective interventions. The identified regions of interest should be closely examined in future studies and clinical practice, potentially leading to more refined early screening methods. Furthermore, understanding how these brain regions are affected by FGR provides a theoretical framework for explaining postnatal developmental delays.

However, it is important to note that this study does not yet provide a clinical tool for immediate use in routine obstetric practice, where FGR diagnosis primarily relies on ultrasonography. Instead, this research highlights the potential long-term neurodevelopmental implications of cortical alterations in FGR-affected fetuses. By anticipating these developmental challenges during the prenatal period, clinicians can better guide parents in preparing for potential delays in milestones such as motor skills, language, vision, memory, and cognition. Our findings offer valuable insights into the focus of early prenatal ultrasonography and provide important references for improving prenatal care, ensuring that infants affected by FGR receive the necessary interventions to support their optimal development.

Limitation and future works

While this study provides valuable insights, there are some notable limitations. First, our analysis is limited to cortical properties, without examining subcortical structures, white matter integrity, or functional connectivity. This limited analysis may overlook essential aspects of FGR’s broader impact on brain development, potentially leading to an incomplete understanding of how FGR affects the entire brain. Second, the lack of longitudinal data prevents us from tracking changes in brain development over time. Although we used data from 27 to 39 weeks of gestation, these data were collected from different individuals, lacking longitudinal consistency at the individual level. Lastly, we only collected maternal age as a demographic variable, excluding other potentially influential factors such as maternal health, nutrition, and socioeconomic status. The absence of these data reduces our ability to fully control environmental influences on fetal brain development.

The developmental trajectory of white matter in the fetal brain undergoes less variation compared to gray matter. Myelination and fiber maturation in white matter are typically delayed relative to cortical gray matter development, with significant changes occurring predominantly after birth [59]. Thus, the focus of this study is on gray matter development during FGR. In future work, we will conduct more comprehensive and detailed analyses of subcortical structures, white matter integrity, and functional connectivity. Additionally, we are actively working on initiatives to gather longitudinal fetal MRI data. Furthermore, incorporating ultrasound image-based cortical reconstruction could provide significant advantages. Ultrasound images are more accessible and can be used more frequently throughout gestation, allowing for collection of more extensive and longitudinal datasets. However, the quality of ultrasound data can significantly affect accuracy of cortical reconstruction, making this process challenging. Despite these challenges, using ultrasound data would provide more dynamic view of brain development, enabling researchers to observe how FGR impacts the brain over time. Additionally, building morphological brain networks based on cortical features could offer deeper insights into structural connectivity and how FGR disrupts brain networks. Analyzing these connections could reveal compensatory mechanisms or identify vulnerable regions, leading to more comprehensive understanding of FGR’s effects and potentially informing early intervention strategies. Moreover, future research would benefit from more holistic research integrating both environmental factors and the effects of FGR on brain development. This could provide a much clearer understanding of how the various factors interact and contribute to the observed changes in the fetal brain.

Conclusion

This study presents a comprehensive analysis of FGR and its profound impact on cortical development, revealing region-specific alterations in cortical properties such as sulcal depth and cortical thickness. Our findings highlight the significant roles of the parietal, occipital, and limbic lobes, suggesting that FGR disproportionately affects areas involved in sensory integration, cognitive processing, and emotional regulation. The identification of both vulnerable and resilient brain regions underscores the complexity of FGR’s neurodevelopmental consequences and emphasizes the need for targeted therapeutic strategies.

By achieving a robust diagnostic performance with key cortical features, we provide a promising foundation for developing reliable cortical-related biomarkers (e.g., cortical thickening in cuneus and temporal pole) for early detection and intervention in clinical settings. Our findings contribute to the growing understanding of FGR’s long-term neurodevelopmental implications, with the ultimate goal of improving outcomes for affected populations. Future work integrating ultrasound-based cortical reconstruction and the development of morphological brain networks could allow for deeper insights into FGR’s impact, offering the potential for more effective prenatal care and early interventions.

Abbreviations

FGR

Fetal Growth Restriction

GW

Gestational Week

GA

Gestational Age

ANOVA

Analysis of Variance

OR

Odds Ratio

AUC

Area Under Curve

HASTE

Half-Fourier Acquisition Single-Shot Turbo Spin-Echo

PACS

Picture Archiving and Communication System

HCs

Healthy Controls

TR

time of repetition

TE

time of echo

FOV

field of view

SD

standard deviation

Authors'contributions

SH: Conceptualization, Methodology, Writing-Original Draft, Writing-Review & Editing. LK: Conceptualization, Writing-Review & Editing. SB: Editing. QX: Editing. KZ: Data Preprocessing. XZ: Data Collection. GC: Editing. MZ: Supervision & Editing.

Funding

This work was supported in part by National Natural Science Foundation of China (grant numbers U23 A20295, 62131015, 82441023), the China Ministry of Science and Technology (STI2030-Major Projects-2022ZD0209000, STI2030-Major Projects-2022ZD0213100), Shanghai Municipal Central Guided Local Science and Technology Development Fund (No. YDZX20233100001001), The Key R&D Program of Guangdong Province, China (grant number 2023B0303040001), and HPC Platform of ShanghaiTech University. Jiangsu Maternal and Child Health Hospital High-level Construction Project (GZL2529).

Data availability

The datasets generated and analyzed during the current study are not publicly accessible due to participant privacy concerns; however, they are available from the corresponding author upon reasonable request.

Declarations

Ethics approval and consent to participate

This retrospective study was approved by the Institutional Ethical Committees of The First Affiliated Hospital of Nanjing Medical University (No. 2021-SR-407). The requirement for informed consent was waived by the ethics committee due to the retrospective nature of the study and the use of anonymized clinical data. All procedures were conducted in accordance with the Declaration of Helsinki and relevant national regulations.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.