The Association between Ventricle Ratio in Preterm Infants and Motor Developmental Delay

The global prevalence of preterm births is increasing, despite a declining birth rate [1]. This trend is particularly evident in developed countries [2]. Numerous factors contribute to the elevated rate of preterm births, including a rise in multiple births due to assisted reproduction and increased maternal age. Advances in perinatal care and neonatal intensive care units have also led to improved survival rates for preterm infants [3, 4]. Although the overall survival rate for premature infants has improved, the rate for low and very low birth weight (BW) infants has significantly increased [5, 6]. These infants are more susceptible to severe disabilities and adverse neurodevelopmental outcomes [7].

Preterm infants face various medical complications and are at a heightened risk for mortality and adverse neurological outcomes, such as developmental delay, cerebral palsy, and intellectual impairment [8, 9]. Disabilities are more prevalent and severe in extremely premature infants. Approximately half of extremely preterm infants (gestational age, GA <25) are reported to have neurodevelopmental disabilities, with another half experiencing severe disabilities when followed up at 30 months in a study of 314 children.

Early diagnosis and intervention are crucial for optimal neurodevelopmental outcomes. Novak et al. [10] demonstrated that neonatal brain magnetic resonance imaging (BMRI) is one of the best tools for predicting cerebral palsy before 5 months’ corrected age. BMRI has a sensitivity of 86–89% for predicting cerebral palsy [11, 12]. Consequently, the combination of functional assessment and BMRI is vital for accurately predicting neurodevelopmental outcomes.

BMRI evaluates structural abnormalities in the brain. It is widely recognized that the prediction of future neurological outcomes should not be based solely on BMRI but should also include neurological examinations [13]. Although BMRI has a high negative predictive value, its positive predictive value for major brain pathologies is only 6.1% [14]. Furthermore, conventional BMRI has difficulty identifying subtle injuries or performing volumetric analysis. Numerous studies have investigated the relationship between early biomarkers extracted from BMRI and future developmental outcomes. However, the outcomes associated with various degrees of abnormalities, or BMRIs without definitive structural abnormalities in prematurely born children, remain debated. Additionally, the developing brains of premature infants are often atypical, making it difficult to predict clinical outcomes using neonatal BMRI. The aim of this study was to evaluate voxel-based brain volumetry in high-risk preterm infants using BMRI at term equivalent age (TEA) and to investigate its association with developmental outcomes, thereby facilitating early diagnosis of developmental delay.

From March 2016 to December 2019, a retrospective review of medical records was conducted on preterm infants admitted to neonatal intensive care units at a single tertiary hospital. The study included infants with a GA of less than 32 weeks, a BW below 1,500 g, a BMRI at TEA ±2 weeks, and a follow-up developmental assessment using the Bayley-III. Those with congenital anomalies were excluded. Infants with definite brain structural anomalies on official BMRI readings were excluded. Readings of BMRI with brain injuries such as hypoxic encephalopathy, periventricular hemorrhagic infarction, ventricular dilatation, or white matter lesions were also excluded. GA was determined by the last menstrual date and confirmed through ultrasonography during the first or early second trimester of pregnancy. Infants with genetic syndromes and/or significant congenital malformations were excluded. Since this study is a retrospective study, informed consent was waived by IRB of Chung-Ang University Hospital (IRB No. 2303-008-19461).

Perinatal and postnatal clinical information of infants, as well as maternal history, was obtained. Birth data, including sex, GA, and BW, were collected. Clinical characteristics of the infants were investigated and included 1- and 5-min Apgar scores, history of respiratory distress syndrome, bronchopulmonary dysplasia severity greater than grade 2, history of sepsis, presence of patent ductus arteriosus, periventricular leukomalacia identified by brain ultrasound, retinopathy of prematurity, and history of seizures. Maternal demographic and clinical data were also reviewed, encompassing maternal age at birth, presence of preeclampsia, gestational diabetes mellitus, and antenatal steroid use.

At Chung-Ang University Hospital, BMRI is conducted at TEA in high-risk preterm infants (GA <32, BW <1,500 g) before discharge from the NICU. The high-risk infants are continuously followed up for their development to allow early intervention for any possible developmental delays. At the corrected age of eight to 9 months, the Bayley Scales of Infant Development, 3rd edition (BSID-III), is performed. The BSID-III was administered by an experienced occupational therapist in the study population. The Bayley-III evaluates cognitive, language, and motor development and is widely used in both clinical practice and research. Age-standardized cognitive, language, and motor composite scores were utilized, with composite scores scaled to a metric featuring a mean of 100 and a standard deviation of 15. The composite scores are derived from various sums of subtest scaled scores. Motor composite scores consist of subtests for gross and fine motor scaled scores [15].

In this study, the motor scales from BSID-III were used to classify groups. Infants with a composite score below 85 were classified into the delayed motor development (DMD) group, while others were classified into the typical motor development (TMD) group. Motor delay is often noticeable during early infancy, and motor dysfunction is also known to be an essential criterion for diagnosing cerebral palsy, for which preterm infants are at high risk [10].

All brain images were acquired using a 3T Philips Achieva scanner with a T1-weighted gradient echo pulse sequence (2,300 ms repetition time, 2.26 ms echo time, 8° flip angle, 256 × 256 square matrix, 180 sagittal slices with 1 mm single slice thickness). Structural images were segmented utilizing ITK-SNAP (v3.8.0, Yushkevich et al. [16]; http://www.itksnap.org/), which offers orthogonal view and 3D mesh generation for manual/semiautomatic segmentation tracing. In brief, (1) T1-weighted MRI scans were visually inspected to exclude significant artifacts, (2) trained researchers manually traced contours of the brain cortex and labeled the inner area on each sagittal slice, (3) labeled volumes were segmented into cerebellum, lateral ventricles, and the remainder, (4) segments were smoothed with a Gaussian kernel to minimize noise.

Despite the advancements in automated brain segmentation methods, the infant brain remains the most challenging subject. Particularly, in infants younger than 6 months, brain images are difficult to distinguish using computational approaches due to the minimal brightness differences between components. Consequently, manual tracing and segmentation continue to represent the gold standard [17]. For infant brain segmentation, two trained researchers visually inspected T1 images and excluded images with indiscernible features (Fig. 1). Both researchers were blinded to the developmental outcomes of the participants. The first researcher primarily focused on precisely outlining the whole brain cortex and visually inspecting the raw images and the 3D meshes constructed from the brain and its segments to ensure overall quality. The second researcher was responsible for inspecting the raw images, rejecting those with critical noises, and outlining the cerebellum and lateral ventricles. Following inspection, manual tracing was conducted using digitizer tablets. Initially, brain cortex contours were drawn, and the inner area was labeled as brain volume. Based on the labeled area in the first step, the cerebellum and lateral ventricles were marked voxel-wise as segments. During the tracing process, segments were continuously adjusted for visual and anatomical accuracy using the 3D mesh model generated in ITK-SNAP.

Upon completion of segmentation, the brain, cerebellum, and ventricles were divided bilaterally. To mitigate noise and potential tracing errors, brain volumes were smoothed with a 6 mm full-width half maximum Gaussian kernel. Given that the segments were relatively small and their voxels occasionally sparse, a 3 mm full-width half maximum kernel was applied for smoothing the cerebellum and ventricles. The smoothing was performed using Statistical Parametric Mapping (SPM12; http://www.fil.ion.ucl.ac.uk/spm/software/spm12/) and MATLAB 9.1.0 (MathWorks, Natick, MA, USA). As the final step, the volume of all components, which is equal to the number of remaining voxels, was calculated. Additionally, the relative proportion of each component was also determined.

To evaluate the demographic and clinical characteristics of infants and mothers between TMD and DMD groups, we performed χ² tests, Fisher’s exact tests, and Mann-Whitney U tests. We compared extracted voxels and ratios within the cerebrum and ventricle between TMD and DMD groups using Mann-Whitney U tests. The ventricle and cerebellum ratio (percentage) were calculated using the following formula: the number of ventricle (cerebellum) voxels divided by the number of cerebrum voxels, multiplied by 100. The motor, cognitive, and language composite scores from the Bayley development assessment were compared between TMD and DMD groups using Mann-Whitney U tests. We compared the numbers of gross motor delay, fine motor delay, cognitive delay, and language delay between TMD and DMD groups using Fisher’s exact tests.

We conducted multiple hierarchical regression analyses with all participants to examine the influence of a discrete set of hierarchical variables on DMD as the dependent variable. To control for multiple comparisons with a small sample size, hierarchical logistic regression analysis, which allows for pooling of data and incorporation of group-level predictors, was used to assess the influence of independent variables on DMD. In modeling the independent variables, 14 variables were grouped into 4 models.

Model 1 incorporated infant demographic factors (sex, GA, BW), model 2 included model 1 plus infant clinical status (Apgar score at 1 min, Apgar score at 5 min, respiratory distress syndrome, bronchopulmonary dysplasia, patent ductus arteriosus, and other medical conditions such as periventricular leukomalacia, retinopathy of prematurity, sepsis, and seizure), model 3 added maternal demographic and clinical characteristics (maternal age, maternal preeclampsia, gestational diabetes mellitus, and antenatal steroid use) to model 2, and model 4 incorporated model 3 plus brain size (ventricle ratio and cerebellum ratio). All statistical analyses were performed using IBM SPSS 24 (IBM^®SPSS, Seoul, Korea), and statistical significance was established at p < 0.05.

A total of 47 infants were enrolled in this study. There were 113 NICU graduates with BMRI at TEA between March 2016 and December 2019 (Fig. 2). Among them, 66 participants were eligible for BMRI analysis. Due to the ventricle being very small and discontinuous in the analysis, there was an issue where the data almost disappeared or completely vanished after smoothing. Even when reducing the smoothing radius to 3 mm, the ventricle in the original data was so small that it continued to disappear. Therefore, those cases were excluded from the analysis. Among the remaining participants, those with BSID-III results and without any exclusion criteria were included in this study.

No significant differences were observed in the demographic and clinical characteristics between TMD and DMD groups. Similarly, no significant differences were found in the demographic and clinical characteristics of mothers with TMD and mothers with DMD groups (Table 1).

Significant differences were identified in ventricle voxels (z = −2.11, p = 0.04) and ventricle ratio (z = −2.11, p = 0.04) between TMD and DMD groups. However, no significant differences were observed in the size and ratio of the cerebrum and cerebellum between TMD and DMD groups (Table 2).

No significant differences were noted in the age of Bayley development assessment between TMD and DMD groups (Table 3). Significant differences were found in the motor (z = 5.00, p < 0.01), cognitive (z = 4.33, p < 0.01), and language composite scores (z = 4.05, p < 0.01) of the Bayley development assessment between TMD and DMD groups. Furthermore, significant differences were observed in the gross motor (z = 4.48, p < 0.01) and fine motor scale scores (z = 4.55, p < 0.01) between TMD and DMD groups (Table 3).

Out of the four models applied in this study, only model 4 was significantly associated with motor development follow-up (Table 4). In model 4, the model χ² (25.241, p = 0.03) and Nagelkerke’s R² (0.673, explaining approximately 67.3% of the dependent variable, motor development delay) indicated that the model was adequate for predicting motor development, with an accuracy of 93.5%. Wald statistics were used to confirm whether each variable had a significant individual relationship with motor development delay. Among all independent variables, only ventricle ratio was a statistically significant predictor of motor development delay (Table 4). None of the four models applied in this study were associated with gross motor and fine motor development, cognition delay, or language delay.

In this investigation, we demonstrated a significant difference in ventricle size and ventricle ratio on BMRI at TEA between TMD and DMD groups. Moreover, when controlling for other factors potentially influencing developmental outcomes, ventricle ratio emerged as a robust, independent predictor of future developmental motor outcomes. Brain volumetry may serve as an adjunctive tool alongside physical and functional examinations for identifying children at risk for developmental motor delays. Early detection and intervention are critical, and our findings may aid in predicting developmental outcomes and expediting diagnosis. Historically, accurate interpretation of BMRI in immature brains has proven difficult, and its precise prognostic value for neurodevelopmental outcomes has been disputed unless assessed in conjunction with other examination tools [10, 13].

Our finding that ventricle ratio, calculated as ventricle voxels divided by cerebrum voxels, is a powerful, standalone predictor of future developmental motor outcomes underscores the importance of brain parenchymal volume for neurodevelopmental outcomes. It has been reported that small fetal brain volume, comprising all parenchymal brain tissue and excluding cerebrospinal fluid, is a strong independent predictor of 2-year neurodevelopmental outcomes and may serve as a valuable imaging biomarker for future neurodevelopmental risk in children with chronic heart disease [18]. Numerous studies have explored the association between brain volumetry in preterm infants and their developmental outcomes. For instance, Peterson et al. [19] discovered that reduced brain volumes, particularly in the cerebellum and hippocampus, correlated with inferior cognitive and motor outcomes at 2 years of age in preterm infants. Thompson et al. [20] found that smaller cortical gray matter volumes were linked to poorer cognitive outcomes at 6 years of age in preterm children. In a study by Kidokoro et al. [21], diminished cerebellar volume at TEA emerged as a significant predictor of motor impairment at 2 years of age in preterm infants. Cheong et al. [22] also reported that larger total brain tissue and cerebellar volumes at TEA were associated with improved neurodevelopment in moderate and late preterm infants. Consistent with previous research, our study supports the association between low cerebral parenchymal volume on BMRI at TEA and adverse neurodevelopmental outcomes in infants.

The size of the ventricles exhibited significant differences between the TMD and DMD groups. Ventricle enlargement is a prevalent brain abnormality in preterm infants, frequently accompanied by cerebral white matter atrophy or aberrant development [23, 24]. As a result, ventricular volume and size serve as proxy measures for assessing the extent of white matter damage. Among preterm infants at TEA, 39–75% exhibit larger ventricles or increased ventricular volumes compared to controls [24, 25]. Fox et al. [26] discovered that larger lateral ventricles in the parietal region at 1 month of age were associated with diminished motor development at 2 years. Increased ventricular dimensions were also linked to delayed early language development. In another study involving very preterm subjects, neonatal ventricular size demonstrated a correlation with more extended neurodevelopmental outcomes at a corrected age of 4.5 years [27]. The authors posited that abnormal white matter maturation may occur in the context of ventricular enlargement, with ventricular dilation itself being linked to a heightened risk of neurodevelopmental sequelae.

The analysis of BMRI at TEA in the absence of significant or major abnormalities is noteworthy, as it implies that even normal BMRI readings can offer valuable information for predicting future outcomes. It has been previously established that accurate outcome prediction is feasible when functional evaluation is combined with brain MRI or ultrasound findings [13]. Prior research has shown that approximately 40% of developmental delay cases exhibited normal BMRI readings [28, 29]. Moreover, Carney et al. [30] reported that incidental abnormalities on neonatal BMRI were identified in up to 47% of participants, with the majority ultimately exhibiting normal developmental outcomes at 18 months. Incidental abnormalities encompassed intracranial hemorrhage, caudothalamic subependymal cysts, acute cerebral infarctions, and punctate white matter lesions.

Several potential explanations exist for the observed relationship between brain parenchymal volume and future neurodevelopmental outcomes. Research has demonstrated that preterm infants with low cerebral volume exhibit reduced cortical thickness and gray matter volume, potentially contributing to cognitive and behavioral development deficits [31, 32]. Moreover, diminished cerebral volume may be linked to compromised white matter microstructure, as evidenced by decreased fractional anisotropy and increased radial diffusivity, possibly reflecting disrupted axonal development or myelination [33, 34].

This study’s limitations include the inconsistent Bayley assessment age across participants, which may have influenced the accuracy of developmental outcomes. Furthermore, manually identifying and measuring ventricles on brain MRI can be labor-intensive and challenging, particularly in cases with small ventricles. Consequently, the feasibility of implementing this approach in clinical practice may be limited. To address these limitations, future investigations might consider employing automated systems, such as deep learning algorithms, to streamline the analysis and interpretation of brain MRI data. Such systems have the potential to enhance the efficiency and precision of MRI interpretation and facilitate easier application in clinical settings. Ultimately, this would expedite early identification and intervention for children at risk of developmental delays.

In summary, the current study, which employed voxel-based brain volumetry in high-risk preterm infants using BMRI at TEA, identified a correlation between reduced brain parenchymal volume and adverse developmental motor outcomes. These findings imply that BMRI volumetry could serve as a valuable instrument for predicting developmental motor outcomes in preterm infants, enabling early interventions and better long-term outcomes.

This article was reviewed and approved by the Institutional Review Board (IRB) of Chung-Ang University Hospital (IRB No. 2303-008-19461), and a written consent requirement was waived. Since this study is a retrospective medical record study, it is a study within the minimum risk in that it does not harm the research subject, and the subject information is anonymized, so the contents that can identify the subject’s identity are not collected, and the personal information is kept confidential.

The authors have no conflicts of interest to disclose.

This study was supported by Basic Science Research Program through the National Foundation of Korea (NRF) funded by the Ministry of Education (2020R1C1C1010486), by the National Research Foundation of Korea (NRF) of the Korean Government under Grant 2018M1A3A3A02065779.

H.I. Shin, D.K. Kim, D.H. Han, and S.M. Kim made study concepts and design. H.I. Shin, D.K. Kim, and N.M. Lee recruited participants. G. Choi, D.H. Han, and H.I. Shin analyzed clinical and brain MRI data. H.I. Shin, H. Hwang, and D.K. Kim prepared and revised the manuscript. D.K. Kim and D.H. Han supervised overall process of study as corresponding authors. The authors had complete access to the study data that support the publication.

All data generated or analyzed during this study are included in this article. Further inquiries can be directed to the corresponding author.