T2* relaxometry to characterize normal placental development over gestation in-vivo at 3T [version 1; peer review: 1 approved, 1 approved with reservations]

Background: T2* relaxometry has been identified as a non-invasive way to study the placenta in-vivo with good potential to identify placental insufficiency. Typical interpretation links T2* values to oxygen concentrations. This study aimed to comprehensively assess T2* maps as a marker of placental oxygenation in-vivo . Methods : A multi-echo gradient echo echo planar imaging sequence is used in a cohort of 84 healthy pregnant women. Special emphasis is put on spatial analysis: histogram measures, Histogram Asymmetry Measure (HAM) and lacunarity. Influences of maternal, fetal and placental factors and experimental parameters on the proposed measures are evaluated. Results : T2* maps were obtained from each placenta in less than 30sec. The previously reported decreasing trend in mean T2* with gestation was confirmed (3.45 ms decline per week). Factors such as maternal age, BMI, fetal sex, parity, mode of delivery and placental location were shown to be uncorrelated with T2* once corrected for gestational age. Robustness of the obtained values with regard to variation in segmentation and voxel-size were established. The proposed spatially resolved measures reveal a change in T2* in late gestation. Conclusions : T2* mapping is a robust and quick technique allowing quantification of both whole volume and spatial quantification largely independent of confounding factors


Introduction
The placenta provides the unique link between the woman and the growing fetus and is therefore crucial for any successful pregnancy.The exchange and transport of oxygen is key to ensure adequate growth and survival of the fetus.Placental insufficiency has been linked to major pregnancy complications such as pre-eclampsia (PE), characterised by maternal high blood pressure and multi-organ disease, and fetal growth restriction (FGR), defined as fetal growth below its potential 1 .The oxygen exchange is enabled by a sophisticated multilevel vascular architecture.Most available assessment techniques focus however, on the summative effect by quantifying flow in the umbilical vein or more recently mean magnetic resonance imaging (MRI)-T2* values across the entire placenta.Moreover, spatial assessment is essential to accurately assess placental function to provide increased understanding, characterisation and quantification of the detailed physiology and thus provide vital information to detect pathological development.

Placental Oxygenation
The exchange of oxygen is enabled by a complex underlying microstructure: The human placenta is composed of multiple lobules, or functional units, defined -following the definition used in previous studies 2 -as the unit of one spiral artery and one corresponding villous tree.The spiral arteries supply oxygenated blood from the uterine arteries into the space between the fetal villous trees, referred to as inter-villous space (see Figure 1A-B).These tree-like heavily vascularised structures originate from the fetal umbilical cord and are bathed in the inflowing maternal blood within the intervillous space (IVS).The transfer of oxygen and nutrients occurs as depicted in Figure 1C-D via a permeable membrane that becomes increasingly thin with advancing gestation.The formation of an adequate vascular system during gestation is essential for the exchange of oxygen and thus placental function.The two-phase angiogenesis with a switch from branching to non-branching angiogenesis is well established from histology 3 .The fetal vascular system develops from initial stem villi, branching out into intermediate and final villi around 27 weeks gestation.These villi contain capillaries which dilate to facilitate gas and nutrient exchange.The oxygen rich blood from the villi in all lobules flows into the umbilical vein which carries the oxygen to the fetus.The placenta insertion point of the umbilical cord varies between central and marginal.
Investigations of the vascular anatomical architecture of the placenta on the lobule level with a range of ex-vivo techniques, in-vivo assessment with ionizing radiation 4 , Power-Doppler Ultrasound and textural analysis of structural MRI scans have led to promising links between structural heterogeneity, fetal growth 5 , villi density 6 and birth weight.
Techniques to assess oxygen concentration include ex-vivo techniques performed immediately after delivery as reviewed by Nye et al. 7 and indirect techniques such as the systemic assessment of angiogenetic factors in maternal blood.They have been linked to pregnancy complications but shows low sensitivity 8 .
Recent years have seen a surge in studies assessing in-vivo human placental oxygenation studies using MRI and mainly T2* relaxometry.This technique exploits the Blood-Level-Dependent effect, relating the concentration of deoxygenated haemoglobin to T2* times.Tissue with lower oxygen concentration thus appears with lower signal in the T2* maps and this can be quantified.In addition to allowing functional oxygenation assessment, T2* relaxometry is an imaging technique, thus allowing access to the macro-structure: While the size of a typical MRI imaging voxel precludes analysis of individual terminal villi (See Figure 1E for a schematic but in-scale depiction of the contained villi to illustrate the length scales), analysis on a lobule level is possible: regions of high T2* surrounded by spheres of lower T2* can be observed and have been shown with dynamic contrast enhanced MRI (DCE-MRI) to correspond to the regions with an early arrival time of contrast.They thus represent the placental regions closest to the maternal inflow 2 .However, despite this clear spatial pattern and studies on rhesus monkeys 2,10 showing promising results, most in-vivo studies to date have reported mean T2* across the whole or from a single region of interest (ROI) within the placenta.This absolute quantification is well suited to illustrate group differences in individual studies, but has the substantial limitation of averaging data and thus loses information about spatial variation which is contained in the T2* maps.This gives huge potential for mis-interpretation.
Quantifying the spatial distribution of T2* over gestation in-vivo thus yields an exciting opportunity for a deeper assessment of the key link between placental function and architecture on a lobule level.This study aims to bridge this gap by providing robust and fast /emphin-vivo techniques able to acquire and quantify spatial effects and by characterising the progression of these over gestation -an indispensable step to detect an MR imaging signature of placental pathology.
We hypothesize, that analysis techniques including texture and spatial distribution in addition to the mean T2* across the entire placenta will inform knowledge of placental development over gestation Table 1 summarizes studies using T2* relaxometry in the human placenta over the last decade.These have varied in field strength between 1.5T and 3T and have employed a range of MRI sequences (Gradient Recalled Echo, 3D-spiral Gradient Echo, Multi-echo Gradient Echo Echo Planar Imaging (MEGE-EPI)), voxel size, number and selection of echo times, coverage, orientation and number of slices and breathing strategy.There is also wide variation in motion correction, data analysis, in the cohorts studied, the gestational age range and in ma ternal position employed.There have also been diverse strategies in image acquisition, such as coverage from one slice to whole volume and coronal to axial scan orientation, corresponding to parallel vs. orthogonal to the maternal-fetal axis, and in the analysis with the selection of specific regions of interest (ROI).Whilst the absolute values obtained differ, there is general agreement on the negative correlation between T2* and gestational age.

Outline
The purpose of this paper is threefold: 1. To present novel quantitative normalized measures which explore the placenta in more depth by studying spatial distribution and texture.
2. To present a comprehensive study of participants with well characterised pregnancies without complications scanned between 20 and 40 weeks' gestation to yield normative curves at 3T.

3.
To study robustness and variability of the proposed measures.

Methods
Cohort with low-risk uncomplicated pregnancies 119 scans from women without known complications at the time of scan, and specifically no diagnosis of GDM (gestational diabetes) PE or FGR were recruited and informed consent was obtained as part of ongoing fetal imaging studies (REC LO/11/1147, 07/H0707/105).Outcome data was collected post scan to ensure their status as 'normal' low-risk All women were scanned on a clinical Philips 3T Achieva with the 32-channel cardiac coil in supine position using a rigorous approach to ensure maternal and fetal comfort and safety: this includes both the preparation and positioning, with women were initially placed in a left lateral and slowly returned to a supine position to keep the uterus and fetus from compressing the vena cava.Women were familiarized with the supine position first and then dedicated padding placed to ensure their comfort before the anterior part of the receiver coil was put in place.Furthermore, continuous monitoring of the maternal heart rate, blood pressure and oxygen saturation was provided for all examinations, with observations taken by either an obstetrician or a qualified midwife during the entire scan.Women listened to their own choice of music but there was regular verbal communication with the radiographer between acquisitions.Lastly, the total examination was limited to 60min in total but split into two sessions, with the women given a break between these sessions.

Data acquisition and processing
A main magnetic field (B0) map was acquired prior to EPI scanning to enable image based (2nd order) shimming 20 .An in-house tool 20 written in MATLAB (2012b; The MathWorks, Inc.) connected to the scanner software allows segmentation of the placenta using predefined ellipsoids, calculates higher-order shim settings and transfers these back to the EPI sequences.Next, a free-breathing multi-slice multi-echo gradient echo (MEGE) echoplanar-imaging sequence was performed.Full placental coverage over gestation was obtained with 20-60 slices.The parameters were optimized to ensure acceptable acoustic noise, with a threshold of 105 dB(A) taken as the cutoff.This effectively increased the required echo spacing in the EPI echo train to reduce the gradient slew rate.Final parameters include resolution (3mm) Voxelwise T2* and proton density maps were calculated from the MEGE-EPI data MATLAB2012b.Since multiple images at different echo times were acquired in a single shot (within <200 msec) for each slice, they are intrinsically aligned and so voxel wise T2* values can be fitted without further processing.No distortion correction or slice removal due to inter-slice motion was performed.A mono exponential decay model was fitted to the measured intensities (S i ) and their respective echo times (T E i ), to obtain T2* maps T2* and proton density maps PD.Non-linear least squares regression was used for the fitting.The image data at the first echo time (S 1 ) provided starting values for PD and the T2* value was initialized with T2*= 100 msec.To speed up the fitting procedure in voxels corresponding to noise, fitting was interrupted after 50 attempts if not converged 14 .Only mono-exponential decay was considered in this study.The number of echoes chosen for the fits was driven by the signal-to-noise ratio such that the last acquired echo had data above the noise floor to avoid Riccian bias.ROIs were drawn by two experienced observers (obs1 and obs2) in MRtrix3 with similar instructions to segment the placenta conservatively, avoiding inclusion of both amniotic fluid and maternal vasculature.The masks were automatically refined by excluding non-physiological values of > 200ms.62 scans with resolution of 3mm were segmented by both observers, 26 of the scans with higher resolution were randomly chosen and segmented by obs1.

Quantification
The mean T2* for each dataset was calculated as the average of both observers, resulting in an average mean T2* 0.5( ) . The T2* maps were transformed into histograms using N b = 60 bins from 0-200, resulting in probability functions P i ϵ [0, 1] N b and bin means M i ϵ R N b .To capture further effects that could be specifically important for the placenta, a simple robust measure reflecting the tail of high T2* voxels in the histogram distributions within the placenta was calculated.This Histogram Asymmetry Measure (H(r)) is defined as the fraction of placental voxels with T2* values greater than a specified ratio (r) of the maximum T2*, defined as the maximal value occurring in at least 1% of the voxels to exclude outliers, over the whole placenta (See illustration in Figure 2A.The effect of varying r was explored by systematic search of values from 0.2 to 4.0. Lacunarity measures were explored to quantify the texture within the imaging data.Lacunarity is a measurement of the spatial distribution of gaps of a specific size.Among many approaches present, the gliding-box algorithm calculates lacunarity by gliding a box of different sizes.Adapted from binary image to grayscale by including the relative height of the box 21 , it was previously used on T2 weighted anatomical images of the placenta 14 , where lacunarity was shown to increase over gestation.Here, lacunarity measures (L(b)) were calculated for different boxsizes b = [3mm, 60mm] to evaluate which length scale is most sensitive to the scale of the placental lobules.By the definition of lacunarity, a tissue which is homogeneously textured at the boxsize level will result in values of 1 and higher values are indicative of heterogeneity at this level.This is schematically illustrated in Figure 2B.
The analysis presented can be performed without Matlab using Octave.

Variability
The mean difference for all values and the Dice coefficient as a measure of agreement in the ROI selection were evaluated.The Dice coefficient is calculated as obs obs DICE obs obs obs obs where | obs1 ∩ obs2 | are voxels both segmenters agree, | obs1 | and | obs2 | are the segmented voxels for observer 1 and respectively.

Covariate analysis
Analysis of covariance (ANCOVA) was performed for all considered values: mean T2* M, volume V, lacunarity L and HAM H to evaluate whether a statistically significant difference exists between the evolution of the parameters over gestation and dependant variables, GA at birth, gestation weeks between scan and delivery, maternal age, maternal BMI and parity, mode of delivery, placental lo cation, fetal sex and birth weight centile.In each case the data was dichotomised for the parameter in question.Similar analysis was also performed with regard to mean T2* in the central and peripheral placental slices, with different resolutions and comparison of the two observers.

Results
An overview over 64 placentas from our cohort is displayed in Figure 3

Spatial analysis
The histograms from all participants are depicted in Figure 7A.They are color-coded by gestational age from yellow (20 weeks) to dark red (41 weeks).A general left shift can be observed in-line with the decrease in T2* with gestation.Furthermore, the width decreases and the fraction at the maximal occurrence (histogram peak) increases from approximately 15% at 20 weeks to 35% at 38 weeks' gestation.In general, a smooth transition can be observed from 20-30 weeks (yellow-orange), before the shape of the histogram changes substantially.
To further quantify these trends, the Histogram Asymmetry Measure H(r) and lacunarity lac(b) were analysed.Given that both measures are parametrized (respectively by ratio r and boxsize b), parameter ranges were studied and the results are shown in Figure 6.The H(r) results (A) remain 1 for small r, such as 0.02, signifying that close to all voxels have T2* values above 2% of the maximal value.Graphically, values below this threshold correspond to voxels at the far left side of the histogram.
As r is increased an increasing number of scans demonstrate H values that are clearly less than one, corresponding to very asymmetric histograms, where a few voxels in the right tail have a substantial effect on the mean, and where the majority of voxels on the left side of the histogram display very low values.It is notable that for moderate r there is a tendency for scans with low H values to have low birthweights, although the converse is not the case (namely low birthweight does not imply low H).The plot for H(0.18), corresponding to a ratio of 18% was chosen to best represent the histogram asymmetry.Lacunarity with a boxsize of 10 voxels, corresponding to 30mm was selected for future use.
These two selected measures are displayed in Figure 7 A-B for all placentas from this low-risk cohort together with examples illustrating placentas with high and low scores in these two measures.

Variability and covariate analysis
The results from the quantitative covariance analysis of mean T2*, H and L and GA focusing on effects of maternal age, maternal BMI, parity, placental location and fetal sex with gestational age showed no significant differences were found for any of these factors, shown by the p-values > 0.05 for any of the evaluated categories.Graphical illustration can be found in Figure 4.
The Dice coefficient was equal to 0.86 between both observers.Significant correlation with the Dice coefficient was shown     for both Mean T2* (p<0.005) and Volume (p<0.005),indicating that differences in segmentations affect these values.However, no significant correlation could be shown for histogram asymmetry score (p=0.803) and lacunarity (p=0.91)-indicating that these are largely independent from small deviations in the segmentations.The analysis of different voxel sizes (2mm and 3mm) did not result in any significant differences in mean T2* (p=0.84),volume (p=0.73),lacunarity (p=0.

Experimental design
The cross-sectional study design covering a wide range of GAs offers considerable benefits compared to studies at fixed GA.These include the ability to obtain robust trend lines similar to previous studies on 1.5T 12,13 , but increases the requirements for the number of participants.
We selected single shot multi echo EPI (MEGE-EPI) with five echo times to acquire all the T2* data.This choice differs from previous studies with a considerably higher number of echo times [11][12][13]15,16,19 , acquired in separate single echoes and GRE [11][12][13]16,19 or 3D spiral read-out 15 . MEGE-EI was selected due to its intrinsic ability to freeze motion during the shot and the close temporal proximity (<200ms) of all data required to obtain T2* per slice.While the acquisition of the required TEs in subsequent single-echoes allows wider choices for the considered echo times, it comes with the drawback of combining data acquired over a larger time window.This is problematic for any motion, but specifically in cases of sudden changes, such as when contractions occur. No respiratry gating or breath holds were used to maximize maternal comfort and to decrease acquisition time. Thse choices led to a total acquisition time of < 30sec per volume, a vendor-available sequence, all of which facilitate widespread application.The chosen 2D approach freezes motion on a slice, but not a volume level.
Two studies so far report within case reproducibility values with generally good results 11,14 .T2* values obtained at 1.5T from healthy uncomplicated pregnancies over gestational age have been reported, 11, showing a decrease of 4.6 ms per week.
Our datasets were acquired in a placental coronal orientation, which differs from studies that have employed a transverse orientation, e.g. the work of Sinding and colleagues 11 .Most placentas have the largest extent in maternal foot-to-head direction (anterior and posterior placentas), so our choice has the advantage that each shot produces a motion-free slice along the longest placental direction.However, a potential disadvantage is that the individual acquired slices do not intrinsically cover the functional maternal-to-fetal direction of the placenta, corresponding here to anterior-to-posterior.
We chose to acquire all datasets in a maternal supine position.This choice was taken to increase scanning efficiency and achieve higher geometric consistency between participants.However, while recently more and more groups acquire their data in this position 14,15 , fetal studies have conventionally acquired data in left lateral tilt (LLT) position.The main aim of the LLT position is to decrease compression of the vena cava from the weight of the pregnant uterus.Our approach includes rigorous continuous monitoring of vital signs, including blood pressure and oxygen saturation by a qualified obstetrician or midwife throughout the scan 22 .Great care is employed to assure maternal comfort, including specially developed padding, time spent on positioning and limiting each scan block to 30 min.All the woman in this study could tolerate the supine position and no episodes of decreasing blood pressure or any other signs suggestive of a aortocaval compression syndrome were recorded.For future studies, attention could be drawn to recent anecdotal evidence that maternal lie might influence quantitative functional placental MRI, e.g. in work by Zun and colleagues 23 .Further work could thus include a systematic analysis of both proposed positions (supine and accurately measured left lateral tilt) for all studied quantitative values.While not affecting the comparison within our study as we scanned consistently in the same standardized supine position, this might affect comparisons with data acquired in left lateral tilt position.

Analysis results
While previous efforts focused on reproducibility 12,14 and showed for the presented MEGE-EPI acquisition good rates of agreements within the same and within subsequent scanning sessions, we included a set of factors and experiments to study further effects.These include systematic analysis regarding fetal sex, maternal BMI, maternal age, parity and placental location over a wider range of gestational ages.However, as mentioned above, the design of this study to include a wide range of gestational ages, comes with the drawback that the studied factors are not equally distributed over this entire range: While we concluded that no significant bias was observed, the sample size calls for caution and a need for further experiments.Specifically for BMI, the maximum value was limited to 35 kg/m 2 in our study due to the limited bore size (60cm) of the employed 3T scanner.
While the observed differences e.g. for ROI selection are small on a group level, the results indicate that complete volume segmentation is warranted, as recent human 14,15 and animal 2,10,24 studies proposed..Our study is in line with previous findings such as the linear decrease of mean T2* over GA and the left shift in histograms.We added texture analysis and histogram asymmetry measure to include the spatial variations into the analysis.The histograms show an ongoing left shift (in-line with the linear decay), but also exhibit a sharp increase in peak value (number of voxels with most frequent value) accompanied by decreased width around 32-35 weeks in line with cited animal studies.The derived Histogram Asymmetry Measure (HAM) captures this observed change and displays non-linear changes around 32 weeks gestation.
We think histogram/texture measures carry two advantages: First, they facilitate inter-organ and inter-time point analysis as they are independent of absolute T2* values which would differ e.g. with employed field strength, thus potentially offering a way forward to compare highly diverse datasets in terms of used scanner and acquisition parameters.This however, depends on whether larger follow-up studies show the robustness and clinical significance of such measures.Second, they allow deeper assessment of for example heterogeneity between patients as a further dimension.Evidence and additional experiments especially in late gestation are required, as these current findings present a first exploration.
The clear association between lower T2* and e.g.FGR as shown in previous studies 12 hints towards a clinical relevance of this measurement.One of its key advantages is the quantitative nature and ability to give absolute and repeatable values.However, no direct association with 'oxygenation' is possible without acquiring more comprehensive data, as factors such as flow, geometry, partial voluming, partial pressure of O 2 , dissolved O 2 in blood plasma and fetal/maternal haemocrit values and the ratio of maternal to fetal blood all influence T2*.The 4:1 ratio of maternal to fetal blood in the placental parenchyma, together with the reduced ppO 2 in fetal blood leads us to speculate that the measured T2* signal is mostly influenced by maternal haemoglobin concentration.The drop of mean T2* across gestation but the conservation of the tail in the histogram with higher T2* values might indicate a greater fraction of deoxygenated blood in the IVS, which might be a consequence of increased demand for O 2 from the growing fetus.The HAM measure and the changes in late gestation in the presented data might be a way of measuring the ratio between supply and increased demand: We speculate that the proposed HAM measure is sensitive to a reserve fraction in the placenta as outlined by Carter et al. 25 : The histogram asymmetry measure is almost 1 up to around week 32, when it drops in most subjects, indicating possibly that the increase fetal demand can not be achieved in the same way beyond this gestational age.The areas of long T2* become increasingly small and a growing share of the placenta is characterized by very low T2*, possibly because no highly oxygenized blood reaches them anymore as it is rapidly taken up or because the tissue is no longer functioning.The HAM data further show a correlation between earlier drop of HAM and lower BW -which could -spaulating further -originate from the fact that the reserve capacity of the placenta is outrun by demand in these pregnancies.
Further important factors of influence on T2* values could be placental calcification 26 known to be diamagnetic, infarcts 27 , haematomas and placental lakes -all affecting the quantitative evaluation.Finally, T2* varies with field strength -hindering its direct use for e.g.absolute cut-off values.Given this complicated multi-factorial origin, measures capturing more then mean T2* over the placenta and providing the ability to assess placentas accross multiple field strengths can be a way forward to a wider use.The observed lacunarity and HAM changes might be able to capture a wider range of placental maturation elements.
The observed increasing lacunarity over gestation quantifies a visual impression of increasing heterogeneity.This is inversely related to the Long Run Emphasis, higher in coarse textures or large homogeneous areas, which was shown to be an important feature in FGR identification 5 .
First, the transverse relaxation time T2*, measured by the employed GE method depends on macroscopic tissue properties, including the macroscopic geometry due to is sensitivity to the variation in diamagnetic susceptibility.The results originate from tissue properties (such as the concentration of haemoglobin) but are influenced by the tissue architecture 28 .Speculating further, within the villous tissue, this might include the distance from the fetal blood in fetal capillaries to the maternal blood in the IVS (depicted in Figure 1B-C).With this distance known to decrease over gestation, a decay in T2* could partly also be influenced by increasingly steep variation of susceptibilites between these two blood compartments.Several other tissue properties that are not related to oxygen concentration can alter T2* such as the T2 of tissue, flow effects and other tissue properties such as potentially the presence of calcifications.
Recent studies 7,29 also draw attention to the effect of -often undetected -both uterine contractile activity and the maternal position as a source of alteration in the T2* values.Uterine contractions have been shown to significantly alter maternal inflow and oxygen concentration in studies of the pregnant ewe 30 .Maternal supine position has been associated with compression of the vena cava due to the pressure of the gravid uterus, particularly later in gestation.A consequent reduced blood supply to the placenta might also decrease T2* measurements.Our proposed fast multi-echo sequence might mitigate the risk to acquire the input for the fits in different contraction states.
This study contributes to the evidence and basics required for future clinical studies.The presented acquisition and derived quantitative measures will in a next step be employed on several cohorts, including pregnancies affected by FGR, PE, GDM and congenital heart disease.A wider application would be facilitated by our proposed un-gated MEGE-EPI sequence -offering quick and uncomplicated acquisition available on all clinical scanners.Important barriers for widescale adoption remain the required manual segmentation of all placental slices.Further work will focus on automating segmentation and quantification and on additional experiments to answer some of the raised questions such as the origin of the changes in the spatial measures around week 32.

Underlying data
The in-vivo MRI data acquired on pregnant subjects under the ethics number [LO/11/1147] and in-line with these ethics we are not allowed to share the imaging data outside of our institution without an agreement in-place.Interested researchers can apply for access to intermediate data, such as processed T2* values, and this access will be granted under certain conditions (academic institution, non-commercial) after an agreement has been put in place.Applications for access should be made via the corresponding author (jana.hutter@kcl.ac.uk).

Open Peer Review
Current Peer Review Status:

University of Manchester, Manchester, UK
A well written paper, with a clearly defined remit which is a significant problem in defining placental T2* in individuals.The problem of heterogeneity within the placenta will be particularly relevant in the characterisation of FGR/PET placentas.This article provides the basis for this type of placental characterisation rather than a mean T2* from a whole placental ROI.

○
The relevant literature is described and the findings of a correlation with gestation is consistent with current knowledge.I note the reference to Ingram et al. in table 1 is incorrect and this should be at 1.5T.Of the results, there are attempts at demonstrating relationships with maternal age and BMI (figure 4); this data whilst interesting is dichotomous and therefore its relevance uncertain.In addition, demographics of this data is not included, is the spread of these parameters consistent across gestation?I think figure 4 could be kept to figure 4.A.

○
This data is complex and therefore the figure captions require more detail, particularly 5, 6 & 7.In figure 6, the results appear the wrong way around from the caption.

○
The methodology suggests some subjects are scanned longitudinally, how is this data incorporated?Have they been included within the cross-sectional analysis multiple times?

Esra Abaci Turk
Fetal-Neonatal Neuroimaging and Developmental Science Center (FNNDSC), Boston Children's Hospital, Boston, MA, USA There is a growing interest in MRI measures of placental function over gestation and this manuscript proposes comprehensive analysis of placental T2* maps, which is necessary before T2* can be used as a diagnostic tool.

Special comments:
Abstract: It would be good to have a summary statement in the 'Results' section related to spatial 1.

Introduction:
Page 3, last paragraph, end of the second sentence: unnecessary word 'times'.1.
Please correct field strength for Ingram2017, which should be 1.5T, and include the gestational age range given for the whole cohort.

Results:
Can you please specify which slice has been chosen to demonstrate the placental T2* maps in Figure 3 (e.g.closer to maternal side/closer to fetal side or just at the center)? 1.
Page 6, first paragraph of results section: I believe the second sentence refers to Figure 4 not Figure 5.

2.
Page 7, first sentence of spatial analysis: I believe this sentence refers to Figure 5 not Figure 7A.

3.
Page 7, second paragraph of spatial analysis: Please check Figure 6 as it is not showing the described information accurately.There are 4 graphs for 8 b values and r values were not indicated for HAM plots.In the figure caption, 'blue frames' should be replaced with 'red frames'.

4.
Page 7, first sentence under 'Variability and covariate analysis'.Please correct the sentence, I think 'were found' has to be removed.

5.
It is unclear to me how you determine the best p and b values giving the best (?) representation for HAM and lac.

6.
Discussion: Page 10, under the discussion of effect of maternal position can you also discuss possible fetal/placental response to the maternal position change?I saw a brief discussion related to that when you discussed the effect of contraction in page 11, but I think for consistency it is better to discuss this before.Discussion related to the change in mean T2* with the differences in segmentation would be good to understand what the main cause would be (e.g.including myometrium?Or excluding some maternal/fetal parts?). 3.

Is the work clearly and accurately presented and does it cite the current literature? Yes
Is the study design appropriate and is the work technically sound?Yes

Are sufficient details of methods and analysis provided to allow replication by others? Yes
If applicable, is the statistical analysis and its interpretation appropriate?I cannot comment.A qualified statistician is required.I confirm that I have read this submission and believe that I have an appropriate level of expertise to confirm that it is of an acceptable scientific standard, however I have significant reservations, as outlined above.

Figure 1 .
Figure 1.Schematic depiction of the human placenta.(A) Model of the human placenta with a zoom into one villous tree in (B).(C)-(D) Schematic drawing of a villous cross-section (as indicated in B) in mid and late gestation, depicting the synciotrophblast and cytotrophoblast layers as well as the stroma with fetal capillaries, Hofbauer cells and fibroblast cells.(E) Illustrates the content of one MRI voxel of size 3 mm 3 , with C/D.A-B not to scale, C-E to scale.[Part A is adapted from 9.] ordered by GA.The mean T2* values from all placental scans at 3T are shown in Figure 5.Using a linear trend analysis, at 3T the mean T2* dropped by 3.45 ms per week to equal 77.13 ms at 24 weeks, 49.54 at 32 weeks and 21.95 at 40 weeks.

Figure 2 .
Figure 2. Explanation of the presented spatial features Histogram Asymmetry and Lacunarity.Thereby, two histograms illustrating a high (1) and low (0.81) Histogram Asymmetry score are shown and the influence of the boxsize (6mm, 30mm and 60mm) is illlustrated for three exemplary cases.

Figure 3 .
Figure 3. Overview of 64 exemplary placentas from the normal cohort sorted by gestational age.

Figure 5 .
Figure 5. Histograms depicting the distribution of T2* values in all voxels within the placenta from all participants.Colored by gestational age at scan, yellow 22 weeks to dark red 39 weeks.

Figure 4 .
Figure 4. Quantitative Mean T2* results for the considered normal cohort.In (A) all samples are shown, (B-F) illustrate mean T2* over gestation for subgroups dichotomized for several maternal and fetal factors.

Figure 7 .
Figure 7. Spatial results from all participants plotted against gestational age for (A) lacunarity and (B) Histogram Asymmetry.Slices with the maximal volume from selected exampled for high/low values are depicted.

Figure 6 .
Figure 6.Results from the (A) HAM(p) and (B) lac(b) measures for different p = 0.04 to 0.20 and b = 2 to 18.The blue frames illustrate the values chosen for further analysis.

Figure 2 :
Figure caption is not clear enough, please clarify A and B in the figure caption.1.

1 .
Page 11, third paragraph: 'The 4:1 ratio of maternal to fetal blood in the ...' -please provide a reference.Does this change with gestational age? 2.
all the source data underlying the results available to ensure full reproducibility?Partly Are the conclusions drawn adequately supported by the results?Yes Competing Interests: No competing interests were disclosed.Reviewer Expertise: I am a researcher at Boston Children's Hospital (BCH) and a member of the Fetal-Neonatal Neuroimaging & Developmental Science Center led by Dr. P. Ellen Grant.My main research topic is fetal/placental MRI and since 2014 I have focused on placental oxygen transport using BOLD and T2* mapping techniques.

Table 2 .
Characteristics of the study participants.The cohort included uncomplicated pregnancies not diagnosed with PE, FGR, GDM and hypertension resulting in a life birth at 36 weeks or above between the 2nd and 98th centile.
Lo et al. and Hirsch et al. paved the way to go beyond that by including e.g.histogram analysis Most previous human studies reported mean T2* as the measure of interest.Previous animal studies fromSchabel et al., This is an open access peer review report distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

work clearly and accurately presented and does it cite the current literature? Yes Is the study design appropriate and is the work technically sound? Yes Are sufficient details of methods and analysis provided to allow replication by others? Yes If applicable, is the statistical analysis and its interpretation appropriate? Yes Are all the source data underlying the results available to ensure full reproducibility? Yes Are the conclusions drawn adequately supported by the results? Yes Competing Interests: 1st
Lacunarity and Histogram Asymmetry Measure are the key interest in this paper as the novel measure of individual heterogeneity and as measures which are not related to absolute T2*.Its therefore of particular interest to the reader.author and I have worked together on an organising committee for an upcoming ISMRM in-utero imaging workshop (Jan -Oxford).I do not believe this has affected my ability to write an objective and unbiased review of the article.
○ ○The author discusses findings and limitations of the work but is rather speculative regarding the HAM changes which occur at 32-35 weeks.○IstheReviewer Expertise: obstetrics, FGR, placental MR, T2* relaxometry, I confirm that I

have read this submission and believe that I have an appropriate level of expertise to confirm that it is of an acceptable scientific standard.
Reviewer Report 22 November 2019 https://doi.org/10.21956/wellcomeopenres.16897.r36939© 2019 Abaci Turk E. This is an open access peer review report distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.