Single vs. multi-slice assessments of in vivo placental T2* measurements

Abstract

Introduction:

Placental health is vital for maternal and fetal well-being, and placental T2* has been suggested to identify in vivo placental dysfunction prior to delivery. However, ideal regions of interest to best inform functional assessments of the placenta remain unknown. The aim of this study is to compare global and slice-wise measures of in-vivo placental T2* assessments.

Methods:

This prospective study recruited pregnant people with singleton pregnancies between December 2017 and February 2022. 3D multi-echo RF-spoiled gradient echo sequences were acquired, and placental T2* values were derived from global and slice-wise approaches. Statistical analyses included Pearson correlation coefficients, concordance correlation coefficients (CCC), intraclass correlation coefficients (ICC), and Bland-Altman analyses.

Results:

Of 115 participants (mean gestational age, 29.25±5.05 weeks), 68 were healthy controls, and 47 were high-risk pregnancies. Global and slice-wise placental T2* assessments for the entire cohort showed no significant difference nor for individual subgroups (healthy controls or high-risk). Pearson correlation values ranged between 0.88 and 0.99 for mean global and slice-wise placental T2*. CCC analyses ranged from 0.88 to 0.99 for mean T2*, and ICC analyses ranged between 0.88 and 0.99 for mean T2*, showing a strong agreement between measurements. Bland-Altman analyses depicted T2* differences across coverage methods, and groups resided within the 95% limits of agreement.

Discussion:

Single-slice placental assessments offer robust, comparable T2* values to global assessments, with the added benefit of reducing post-processing time and SAR exposure. This supports slice-wise approaches as valid alternatives for assessing placental health in various pregnancies.

Introduction:

Placental health is vital for ensuring the optimal well-being of both the fetus and the mother, yet direct in-vivo assessments remain limited; hence, placental dysfunction is typically suspected only after maternal-fetal health is compromised. [1] Magnetic resonance imaging (MRI) has emerged as a valuable non-invasive technique for evaluating in-vivo placental function, with T2* imaging providing valuable insights into placental oxygenation and potential abnormalities. [2] Previous studies have demonstrated significant differences in T2* values of the in vivo human placenta in pregnancies complicated by fetal growth restriction and small for gestational age infants, highlighting the potential of T2* imaging in detecting abnormalities and predicting adverse pregnancy outcomes prior to delivery, and thus opening critical windows of intervention. [3–6]

However, prior to the clinical implementation of T2* measurements of the placenta, several technical tasks need to be addressed. Automated placenta segmentation remains challenging [7,8], and while emerging techniques continue to refine this technology, manual segmentation remains the current gold standard. Manual segmentation is time-intensive and user-dependent and thus may influence the accuracy and reproducibility of selected regions of interest (ROIs). Furthermore, no consensus exists on using slice-wise or global placental ROIs for T2* assessments. Some studies utilize slice-wise approaches [3–6] focusing on specific areas of the placenta, while others measure global T2* values across the entire placenta [9,10]. While global assessments may better capture the heterogeneity and variability within the placenta, these segmentations are more time-consuming and may be more prone to inaccuracies by contamination from surrounding tissue. While slice-wise ROIs may overcome these limitations, it remains unclear if limiting the ROI significantly alters T2* measures and thus reduces the importance of information gained about placental health and function.

To address these challenges, our main objective was to compare the performance of two approaches, namely global multi-slice versus slice-wise placental assessments, on in vivo T2* values. Additionally, we investigate whether these approaches perform similarly for both healthy and high-risk pregnancies.

Methods:

This study was approved by the institutional review board at the Children’s National Hospital, Washington D.C., US. All study participants provided written informed consent before enrollment and study procedures. Pregnant persons were recruited from December 2017 to February 2022 as part of a prospective observational study. Inclusion criteria included (a) maternal age ≥ 18 years and (b) with a singleton pregnancy. Exclusion criteria included (c) fetuses with known chromosomal abnormalities or genetic syndromes and (d) individuals who did not meet standard criteria for MRI safety. Pregnant participants underwent placental MRI up to two-time points in gestation: the second and/or third trimester.

Relevant clinical data were extracted from the medical records. Subjects were categorized as healthy controls or high-risk (Table 1). The control group consists of pregnant individuals with no significant maternal conditions, medications in pregnancy, or abnormal screening results throughout gestation. The high-risk group comprises those with significant pre-pregnancy or gestational conditions such as chronic hypertension, gestational hypertension, pre-eclampsia, COVID-19 infection, or fetal growth restriction (Figure 1).

Table 1.

Study Population Demographic and Clinical Characteristics

Parameter	All (N=115)
Gestational Age at MRI (Weeks)
Mean (SD)	29.3 (5.1)
Maternal Age at MRI (Years)
Mean (SD)	34.0 (5.0)
Fetal Sex
Male	45 (40.54%)
Female	66 (59.46%)
Birth Weight (Grams)
Mean (SD)	3205 (585)
Gestational Age at Birth (Weeks)
Mean (SD)	39.0 (1.9)

Notes: Number of samples (N), standard deviation (SD), and the p-value for the difference between control and high-risk based on the t-test for continuous variables and the Fisher exact test for categorical variables.

Figure 1:

Flowchart of included subjects. Note: The high-risk group included chronic hypertension (n=8), gestational hypertension (n=8), pre-eclampsia (n=8), fetal growth restriction (n=14), COVID-19 exposure (n=5), or more than one high-risk feature (n=5).

MRI scans were performed without sedation or contrast A 3D multi-echo RF-spoiled gradient echo (GRE) sequence with flow compensation in both readout and slice-encoding directions and a matching T2-weighted (T2-W) 2D single-shot fast spin echo was acquired. Prospective respiratory gating was employed based on respiratory bellows and a 30% acquisition window at end-expiration. All GRE sequences were acquired with whole placental coverage based on the median slice of the placenta unique to each placental position. The slice prescription was adjusted to capture the complete tissue while the participants breathed normally. Imaging was conducted on a GE Discovery MR450/MR450w 1.5T scanner using an 8-channel cardiac coil. For 3D GRE, imaging parameters were FOV = 36–38 cm, acquisition matrix size = 256 × 128, slice thickness = 3 mm, flip angle = 30°, TR = 93 to 97 ms, 15 to 16 TEs ranging from 4 to 80 ms, and pixel bandwidth = ±31.25 kHz. The total scan time with SENSE (R = 2) was 6 to 8 minutes. For reference anatomical imaging, T2-W 2D single-shot fast spin echo imaging was acquired with matching coverage, with the scan parameters of TR/TE = 1200/160 ms, matrix size = 256 × 192, FOV = 36–42 cm, and slice thickness = 3 mm.

Placental segmentation was performed on GRE multi-echo data using the ITK-SNAP software (version 3.8.0), guided by T2-W anatomic images. T2* values were derived from GRE data through voxel-wise fitting of a monoexponential function, with noise-restricted voxels set to 400 [11]. Placental T2* values were computed for global and slice-wise placental coverage, with the latter achieved through stepwise decrements in placental coverage by selecting 11, 9, 7, 5, 3, and one slice(s), maintaining continuous coverage from 33 to 3 mm across the placenta (Figure 2). This approach aligns with prior studies encompassing 10 – 18 mm placental ROIs (9–12). When using a single slice, the central slice was preferred. If there were an even number of slices in global placental segmentations, the central slice was randomly chosen from the two central slices. For the multi-slice analyses, we then selected continuous slices centered around the chosen central slice(s) to ensure comprehensive and representative coverage. To ensure the central slice was the most comprehensive for the slice-wise approach, we calculated the average number of pixels per selected slice to ensure it reflected the highest number of relevant pixels (Supplementary Figure 1).

Figure 2.

Representative figure of global and multi-slice placental slices. This figure showcases global and multi-slice placenta slices for a pregnant participant at 27.14 weeks gestation. The number of slices for the subject was 30, with 20 slices covering the entire placenta. The figure demonstrates placental coverage in the sagittal plane; advancing columns in each row represent decreasing coverage, (A) global coverage in the first column, (B) 11 slices in the second column, and (C) 5 slices in the third. The blue lines represent the Field of View, indicating the area captured and visible in the imaging. The red lines indicate the positions of slices, demonstrating the specific sections or layers of the scanned area. The green colors outline the Region of Interest dedicated to the placenta.

Descriptive and exploratory analyses were applied to summarize and characterize participants in the study. Continuous variables were described using means and SD. Categorical variables were presented as counts and percentages. The student’s t-test was used to evaluate group differences. Pearson correlations were computed to examine the association between approaches using mean T2* values. The intraclass correlation coefficients (ICC), concordance correlation coefficient (CCC), and the Bland-Altman method were used to assess the agreement between the measurements. All p values were two-tailed, with p ≤ .05 considered statistically significant, and confidence intervals were computed at a 95% confidence level. All analyses were performed using MATLAB 2020b, MathWorks, Natick, Mass [12].

Results:

A total of 130 pregnant participants underwent placental MRIs. The quality of MRI data was assessed; 15 examinations were excluded from the placental analysis because of severe MRI artifacts. Ultimately, 115 subjects were included, 68 healthy controls and 47 high-risk individuals. Among the participants, 21 healthy controls and six high-risk individuals had data available at two different time points. A flowchart shows the selection of study subjects from the enrolled pregnant participants Figure 1. The mean gestational age at MRI for the entire cohort was 29.25±5.05 weeks. Global placental segmentations included a mean of 24 slices (SD ±7; range 12–56 slices across gestation). Table 1 Provides a summary of the clinical characteristics of the study cohort. All segmentation was done manually by one of our team members (M.P., L.HC., and K.K.). Thirty-three scans were randomly chosen and segmented by the alternate examiner to evaluate the Interrater reliability (IRR). IRR was higher than 99%.

The mean T2* value for the entire cohort was 101.62 ms using the global approach and the mean T2* value using the single-slice approach was 102.85 ms (Table 2); for healthy controls, the mean T2* using the global approach was 107.12 ms compared to 108.64 ms for the single-slice approach, and 92.38 ms (global) vs. 93.11 ms (single-slice) for the high-risk population. A comparison of mean T2* values using global and decreasing slice-wise measurements did not significantly differ across the entire cohort (Figure 3) nor in control or high-risk groups when analyzed separately (Figure 4).

Table 2.

Comparative analysis of placental mean T2* values using global and slice-wise ROI approaches across all subjects, control, and high-risk groups.

Parameter	All (N=142)	Control (N=89)	High-risk (N=53)	p-Value
Mean T2* in msec (SD)
Global ROI	101.62 (32.21)	107.12 (30.41)	92.38 (33.32)	0.01
11 slices ROI	102.72 (33.09)	108.32 (30.80)	93.31 (34.93)	0.01
9 slices ROI	102.90 (33.68)	108.51 (31.34)	93.47 (35.63)	0.01
7 slices ROI	103.27 (34.52)	109.10 (32.91)	93.49 (35.27)	0.01
5 slices ROI	102.72 (32.98)	108.29 (30.75)	93.37 (34.74)	0.01
3 slices ROI	102.70 (33.24)	108.41 (30.87)	93.13 (35.14)	0.01
Single slice ROI	102.85 (34.00)	108.64 (31.96)	93.11 (35.37)	0.01

Notes: Number of samples (N), Region of Interest (ROI), and standard deviation (SD) and comparison between groups used t-test analyses.

Figure 3.

Comparison of Mean T2* of the placenta. Mean T2* values for global and multi-slice placental coverage approaches across the dataset (n=115). Each dot represents an individual data point. Boxplots summarize the distribution of a dataset using the minimum value, the first quartile (25th percentile), the median (50th percentile), the third quartile (75th percentile), and the maximum value. Note: (G) stands for global placental coverage, (R11) stands for multi-slice placental coverage with 11 slices, (R9) stands for multi-slice placental coverage with nine slices, (R7) stands for multi-slice placental coverage with seven slices, (R5) stands for multi-slice placental coverage with five slices, (R3) stands for multi-slice placental coverage with three slices, and (R1) stands for placental coverage with one slice.

Figure 4.

Comparison of Mean T2* of the placenta. Boxplots and scatter plots are implemented to represent the comparison of mean T2* values across control (A) (n=68) and high-risk (B) (n=47) groups between the global and multi-slice placental coverage. Each dot represents a single T2* value. The box in a boxplot summarizes the distribution of a dataset using the minimum value, the first quartile (25th percentile), the median (50th percentile), the third quartile (75th percentile), and the maximum value. Note: (G) stands for global placental coverage, (R11) stands for multi-slice placental coverage with 11 slices, (R9) stands for multi-slice placental coverage with nine slices, (R7) stands for multi-slice placental coverage with seven slices, (R5) stands for multi-slice placental coverage with five slices, (R3) stands for multi-slice placental coverage with three slices, and (R1) stands for placental coverage with one slice.

Pearson correlation coefficients indicated strong positive relationships between mean T2* values for global and slice-wise placental approaches, with correlation values ranging from 0.88 to 0.99 (Table 3). CCC values are also strongly aligned with each group’s global and slice-wise placental methods, ranging from 0.88 to 0.99 (Table 3). ICC further confirmed a robust agreement between global and slice-wise approaches, ranging from 0.88 to 0.99 (Table 3).

Table 3,

Comparing Mean T2* Values of the Placenta. This table presents a comparison of mean T2* values of the placenta across different coverage approaches. The data is categorized into three groups: All participants, the control group, and the high-risk group. Pearson correlation coefficient, concordance correlation coefficient, and Intraclass correlation coefficients are provided for each comparison.

Parameter	All	All (N=142)	All	Control	Control (N=89)	Control	High-risk	High-risk (N=53)	High-risk
Mean T2* of the placenta
	PCC	CCC	ICC	PCC	CCC	ICC	PCC	CCC	ICC
G vs. R11	0.97	0.97	0.97	0.96	0.96	0.96	0.99	0.99	0.99
G vs. R9	0.97	0.97	0.97	0.95	0.95	0.95	0.99	0.99	0.99
G vs. R7	0.96	0.96	0.96	0.94	0.94	0.94	0.98	0.98	0.98
G vs. R5	0.95	0.95	0.95	0.93	0.93	0.93	0.98	0.97	0.98
G vs. R3	0.93	0.93	0.93	0.91	0.91	0.91	0.96	0.96	0.96
G vs. R1	0.92	0.92	0.92	0.88	0.88	0.88	0.96	0.96	0.96

Notes: Number of samples (N), Pearson correlation coefficient (PCC), concordance correlation coefficient (CCC), intraclass correlation coefficients (ICC), Global Coverage of Placenta (G), multi-slice placental coverage with 11 slices (R11), multi-slice placental coverage with nine slices (R9), multi-slice placental coverage with seven slices (R7), multi-slice placental coverage with five slices (R5), multi-slice placental coverage with three slices (R3), and single-slice coverage of placenta with one slice (R1).

Bland-Altman analysis revealed minimal variations in mean and SD T2* differences between groups across different coverage methods, and most fall within the 95% limits of agreement, indicating overall consistency between the approaches (Supplemental Figure 2).

Previously, it has been shown that mean T2* values decreased with advancing gestational age (GA) at MRI. We confirmed this association in this cohort and further show that this relationship remains significant for global and all slice-wise mean T2* measurements (ranging from r = −0.68 to −0.63, all p <.01) (Supplemental Figure 3).

The study cohort includes subjects with MRI data available at one or two distinct time points. A notable subset of this cohort, comprising 27 subjects, had MRI data available from two separate time points (interval range 4 to 18.9 weeks). The primary analysis was conducted on all available MRI data from the entire cohort. To verify the robustness and sensitivity of the results, the study further categorized the data into four distinct groups (Supplementary Table 1): Model 1: One scan per subject, utilizing the first available scan for subjects with two scans. Model 2: One scan per subject, utilizing the second available scan for subjects with two scans. Model 3: One scan per subject, utilizing a randomly chosen scan (either the first or the second) from subjects with two scans. Model 4: Featured the opposite scan of the one randomly chosen in Model 3 (if the first scan was chosen for Model 3, then the second scan was used for Model 4, and vice versa), in addition to subjects with only one scan. No significant differences in the results were observed across all these different analytical approaches (the entire cohort and the four subgroups).

Discussion:

This study sought to evaluate the comparability of slice-wise and global placental coverage assessments in terms of in vivo T2* values. Our data revealed statistically insignificant differences between mean T2* single slice and global placental assessments. Importantly, the performance of single-slice placental coverage assessments remained consistent across healthy and high-risk pregnancies, suggesting that these approaches are robust, efficient alternatives for interrogating in vivo placental function.

Despite the increasing number of studies of placental T2* to assess placental function, post-processing approaches remain inconsistent across studies, limiting the generalizability needed to import these assessments into routine clinical practice [7,8]. Benefits of single-slice assessments include reduced acquisition times, decreased variability from manual segmentations, and reduced post-processing time. Slice-wise approaches reduce the number of slices and field of view for T2* acquisition, thus decreasing acquisition times, Specific Absorption Rate (SAR), decreasing the acoustic load on the mother, and enhancing the comfort and the safety of the MRI procedure for both the fetus and the pregnant mother [13]. Direct measures of SAR are influenced by both MRI parameters and patient-specific factors such as weight and body habitus, making individual SAR measures complex; however given that SAR is directly proportional to the imaging flip angle and inversly proportional to the RF pulse duration and TR of the sequence, decreasing the number of slices would decrease number of RF pulses applies and thus decrease SAR exposure. Furthermore, manual segmentations remain the gold standard until better automated placental segmentation pipelines become available so that slice-wise approaches require shorter post-processing times. Slice-wise approaches are collectively better suited for large-scale studies and clinical applications. Given these benefits, slice-wise approaches are preferred, presuming important information on placental function is not lost compared to comprehensive, global assessments, as shown in this work.

We demonstrate high correlation coefficients observed across global versus slice-wise approaches. Though statistically insignificant, correlation coefficients do increase minimally with the number of slices used. Despite this increase, additional information learned is minimal, so the single-slice ROIs can effectively capture meaningful T2* values, especially when centered within the placenta. Centralized assessments capture the most metabolically active regions of the placenta, with increased vascularity for efficient gas exchange compared to the peripheral regions, and thus are more likely to reflect placental higher oxygenation [14] and higher mean T2* values [14–16]. Similar to previous studies [3,5,11,17–19], we show decreasing mean placental T2* measurements with advancing gestation, and this relationship is consistent between the global and single-slice approaches.

Despite the strengths of this work, there are several limitations to consider. First, all segmentations were manually performed, resulting in user-induced variability. However, this variability may be overcome with appropriate training, as demonstrated by the high IRR in this report. Placental acquisitions may be affected by maternal and fetal motion. This effect, however, may be minimized with single-slice approaches and shorter acquisition times. Efforts to optimize the segmentation process and ensure accurate placental coverage are necessary to enhance the validity of our findings across platforms and studies. We also show incremental differences in mean T2* values with increased slice ROI, with lower T2* values of global approaches. These differences are statistically insignificant, and several studies suggest an inherent redundancy in placental tissue, especially within the periphery, where placental infarcts are common and not necessarily suggestive of pathology [20]. Further study is needed to fully understand how the periphery of the placenta contributes to placental function. Similarly, future studies should consider the potential clinical significance of central vs. slice-wise placental function and thus drive the relevant approach (single, multi- or global) and subsequent interpretations. Another limitation of our study is the treatment of individuals who underwent two distinct scans at different time points as independent data entities. While this approach may introduce bias in the data analysis, the statistical approach of sensitivity analyses increased the validity and robustness of our findings. Our study utilized slices coronal to the placenta. Different slice orientations could potentially yield varying T2* measurements due to differences in placental tissue composition and oxygenation levels. Further research is needed to explore these variations. Additionally, while this work was applied to high-risk maternal conditions, structural abnormalities of the placenta or umbilical cord, including marginal, eccentric, or velamentous cord insertion, were excluded from this analysis. The accuracy and relevance of this approach to pregnancies complicated by structural abnormalities of the placenta and cord, would be an important future study.

In conclusion, our study provides valuable insights into the comparison of global and slice-wise placental coverage for assessing placental T2* value. We found no significant differences in mean T2* measurements between the single-slice approaches and global placental coverage for both healthy and high-risk pregnancies. Given the known and expected increase in placental heterogeneity in high-risk conditions, this work supports the application of single-slice approaches across both groups. This work also suggests that utilizing placental T2* values with single-slice placental coverages is feasible and may offer advantages regarding reduced acquisition and post-processing times, furthermore reducing SAR. Ongoing efforts to optimize imaging protocols across field strengths and imaging platforms and the reliability and automation of segmentation techniques will continue to improve the accuracy and reliability of placental T2* measurements. Thus, single-slice placental assessments of T2* may serve as a promising biomarker in future large-scale studies of placental function.

Highlights:

• There are minimal differences in T2* measurements between regional and global placental approaches.

• High performance of both regional and global T2* placental assessments apply to both healthy and high-risk pregnancies.

• Regional placental assessments offer T2* values with reduced post-processing time and SAR exposure.

List of Abbreviations:

MRI

Magnetic Resonance Imaging

ROI

Region of Interest

GRE

Gradient Echo

T2-W

T2 Weighted

IRR

Interrater Reliability

SD

Standard Deviation

ICC

The Intraclass Correlation Coefficients

CCC

Concordance Correlation Coefficient

GA

Gestational Age

SAR

Specific Absorption Rate