Two‐Compartment Perfusion MR IVIM Model to Investigate Normal and Pathological Placental Tissue

Perfusion and diffusion coexist in the placenta and can be altered by pathologies. The two‐perfusion model, where f1 and, f2 are the perfusion‐fraction of the fastest and slowest perfusion compartment, respectively, and D is the diffusion coefficient, may help differentiate between normal and impaired placentas.

surrounding the intervillous space, which constitutes the maternal side.The chorionic plate is characterized by the villous trees that are vascularized with fetal blood and exchange nutrients and substances with the surrounding maternal free blood that flows through the intervillous space. 1 The placenta may incur structural and physiological abnormalities such as intrauterine growth restriction (IUGR) and placental accretism. 1he placenta accreta is characterized by the absence of the maternal decidua, so the villous trees can proliferate and anchor directly on the myometrium. 1 There are three degrees of abnormal placental infiltration: the placenta accreta where villi do not invade muscles, the increta where villi invade the myometrium and the percreta where villi cross the serous membrane and can infiltrate the bladder or the rectus depending on the position of the accretion zone. 1 The accretion zone is highly perfused, so this pathology could cause hemorrhages during the delivery and could also lead to a hysterectomy.2 As placental function is related to the health of the newborn and the future adult, 3,4 the development of noninvasive diagnostic techniques to assess and monitor it is desirable.
IUGR is due to the anomalous trophoblastic invasion of the spiral arteries 1 : the arteries are narrower than in normal subjects, and this causes the maternal placenta's incoming blood to have higher pressure and velocity. 1Because the fetalmaternal blood exchange occurs at the interphase between the intravillous maternal placenta and the villous membrane (the trophoblast), the oxygenated blood could be more drained than in healthy placentas causing a decrease of the trophoblastic functionality. 1 IUGR fetuses are characterized by an estimated fetal weight (EFW) below the 10th percentile, so they are smaller than healthy normal fetuses and they could develop heart and neurological disease after birth. 5,6IUGR fetuses may be differentiated into fetal growth restriction (FGR) and small for gestational age (SGA) groups, according to the presence or absence of fetoplacental Doppler abnormalities detected in utero.However, while ultrasound is currently considered the primary diagnostic tool to predict perinatal outcome, [6][7][8] its imaging and flowmetry cannot assess the micro-perfusive and microstructural placental qualities.MRI may be employed as an alternate method to investigate the placental tissues. 9In particular, diffusion weighted imaging (DWI) is a powerful technique that provides microstructural information without requiring contrast agents that could result in adverse effects on fetal development. 10iven the complexity of biological tissues in which perfusion and diffusion compartments coexist, different models have been developed to approximate the DWI signal. 11The most widely known and used is the intravoxel incoherent motion (IVIM) model, a bi-exponential model that considers two separate compartments: that of perfusion quantified by the perfusion fraction f IVIM of perfusing molecules at a rate given by the pseudo-diffusion coefficient D*, and the diffusion compartment quantified by the diffusion coefficient D 12 of 1 À f IVIM water molecules.4][15][16] In the placental tissue, f IVIM quantifies the perfusion fraction of water molecules perfused in microcapillaries with D* rate, whereas D, which quantifies the hindered diffusion of water molecules in the extracellular space, is related to tissue microstructure.Some authors have identified f IVIM as a biomarker to discriminate between IUGR and normal fetuses [13][14][15] and to discriminate between normal and accreta placentas. 16In this study, we hypothesized that, in the placenta, two main perfusion compartments exist in addition to the diffusion compartment, and that the introduction of more parameters describing placental perfusion can provide more information to identify the placenta's physiological and microstructural characteristics and understand the mechanism involved in placental diseases.
Thus, the aim of this study was to investigate the potential of a three compartment (two perfusion and one diffusion) model, based on the two-perfusion model developed by Fournet et al.,17 to discriminate between normal, SGA, FGR, and accreta placentas.

Materials and Methods
Fig. 1 shows a schematic representation of the pipeline used in this study.

Study Cohort
Following the study's approval by the ethical committee of the Policlinico Umberto I, Sapienza, Rome, Italy, 85 singleton pregnancies with an average gestational age (GA) of 20 AE 2.5 weeks (15-27 weeks) were enrolled from January 2018 to March 2022.All patients completed written consent forms prior to the study.One patient was excluded as it had not a singleton placenta, 19 other patients were excluded because of motion artifacts and signal noise ratio (SNR) lower than 3 a.u.for high b-values.The final study subjects were comprised of 65 patients who were divided into four groups: normal pregnancy (Normal), n = 43; FGR, n = 9; SGA, n = 6, and accreta, n = 7.The accreta group consisted of n = 4 accreta, n = 1 increta, and n = 2 percreta (Table 1).

Model
Placenta physiology is characterized by at least two main perfusion compartments: the compartment given by the exchange of substances through the trophoblastic cells between the mother and fetus side and the perfusion compartment related to pumped blood inside the villous trees.It is reasonable to think that the perfusion rate of the two compartments is different by at least one order of magnitude. 18oreover, the diffusion compartment in the placenta is mainly due to the blood flowing in the intravillous space.
The model described by Fournet et al. 17 considers the contribution of two vascular pools (capillaries and larger vessels) in rat brains: The two-perfusion model foresees the presence of diffusion in each compartment and divides perfusion compartments into slow and fast perfusion.We adapted Fournet et al. model 17 to study the placenta tissue adding the slow perfusion contribution to the first fast perfusion compartment: where D is the pseudo-diffusion coefficient relating to the maternal blood flowing inside the intravillous space, D * 1 is the fastest pseudo-perfusion coefficient given by the blood  flowing inside microvessels and villous trees, D * 2 is the slower pseudo-perfusion coefficient given by the exchange of nutrients between the maternal and the fetal compartments, f 1 is the fastest perfusion fraction and f 2 is the slowest perfusion fraction related to the trophoblastic cells.The choice to add D * 2 in the fast perfusion compartment was because it could not be distinguished from the fastest D * 1 relating to the villi's vasculature, thus D * 2 should contribute to the actual fast perfusion compartment.
To reduce the number of free parameters, the diffusion coefficient D was estimated by fitting a mono-exponential model at high b-value b ≥ 200 s m 2 À Á to DWI data.Since the placenta has two well-defined perfusion processes contributing to the global perfusion, the IVIM model was used to obtain the total perfusion fraction Therefore, the resulting model used in this work is: where f IVIM and D mono (the D estimated by a mono-exponential model) are fixed.All the models were fitted to DWI data using a homemade Python script using a nonlinear leastsquares algorithm.

Data Acquisition
DWIs were acquired using a 1.

Preprocessing
The maternal and fetal sides of placentas were analyzed.
Given the complexity of the placental tissue, six regions of interest (ROI) were manually delineated on the fetal and maternal sides by two different radiology specialists both of 5 years of experience (all results relating to these six regions are available in the Supplemental Material).In accreta placentas the ROI was delineated on the accretion zone.
In multiple-coil acquisition, the noise is known to follow a noncentral χ distribution, which collapses in a χ distribution on the background with a number of degrees of freedom relying on the coils' number. 19The noise was estimated using the following estimator based on local moments 19 : where x is the local mean of the corrupted squared signal calculated using a local window size 21 Â 21, b σ 2 L is the estimated squared noise of the image, and L is the number of coils.In clinical scanners, row data are already filtered by default with a weak filter, which is of an unknown type because it is protected by manufacturing company licenses.Hence, the exact noise distribution is unknown.In this study, therefore, the "mode" has been approximated as the maximum of the signal intensity histogram, and the resulting corrected signal is given by 19 : where the corrupted signal's second order M 2 L x ð Þ x was calculated over every single ROI for the analysis of the tissue.Diffusion and perfusion parametric maps were obtained by first performing a voxel-wise signal correction, where x was estimated by implementing a 2 Â 2 filter on each voxel (example shown in Fig. 2).Then, the twoperfusion IVIM model was fitted using a bugged trees algorithm provided by MATLAB's machine learning official toolbox (MATLAB R2021a).In particular, the bugged trees are built by the function TreeBagger(), then the function predict() was used to obtain the maps.

Fitting Controls
Due to the number of parameters that need to be estimated using the two-perfusion IVIM model, there is a danger of overfitting.To avoid this and to compare IVIM and the twoperfusion IVIM models, the Akaike information criterion (AIC) 17,20 was applied corrected for small size samples.AIC is defined as: where N b is the number of b-values, MSE is the mean squared error, and k is the number of the model's parameters.As shown by Riexinger et al., 21 the comparative goodness of a model can be evaluated by considering the difference between the two competitor models' AICs (eg, AIC 2perf À AIC IVIM Þ.A negative value indicates that the first model (in the example, two-perfusion) is more suitable for describing that data.The following AIC differences were calculated: AIC 2perf À AIC IVIM ¼ À5:53 and IC 2perf À AIC mono ¼ À23:85, the negative values indicating that the two-perfusion model is more suitable than the IVIM and mono-exponential models.The Akaike weights 22 were calculated for each model as: Þ is proportional to the likelihood of the ith model given the data L M i jData ð Þ , Akaike weights can be considered as the probability that the ith model is the best model given the data and the set of models.Indeed, as shown by Fournet et al., 17 a model's Akaike weight higher than the threshold 0.9 indicates that the model may be considered the best model of the set, and a robust inference may be possible.The Akaike weights 22 were calculated for each model (mono-exponential, IVIM, and two-perfusion) with that of the two-perfusion model being best (w 2perf AIC ð Þ¼ 0:94; Fig. 3).

Statistics
Continuous variables were expressed as mean AE standard deviation.All the parameters' values obtained from each placenta group were analyzed by performing a Cohen's d test 23,24 and an ANOVA test with Dunn and Sid ak's posthoc correction (MATLAB 2021a).Since the ANOVA required the groups' homoscedasticity, a Levene test was performed to confirm the null hypothesis of equal variances across the groups.Due to the small number of placentas with different accretism, to evaluate the results obtained in accreta, increta, and percreta compared to healthy placentas, the Cohen's d effective size was used.
Regarding the correlation analysis, Spearman's coefficient was evaluated.A P-value <0.05 indicated a statistically significant difference or correlation.

Examples of Two-Perfusion and IVIM Maps
Three slices from the parametric two-perfusion and IVIM maps of an example healthy placenta with GA = 22.4 weeks are shown in Fig. 4. In the first slice in Fig. 4a, the maternal side is outlined in blue, whereas the fetal side is in green; the red color outlines the umbilical cord and its insertion in the second and third slices.All the perfusion fractions f IVIM , f 1 and f 2 maps (Fig. 4d,b,c, respectively) had higher values in the region of the umbilical cord insertion, and in the decidua (i.e., the maternal side), which is highly perfused by the spiral arteries.The perfusion coefficient D * 2 in Fig. 4g showed patterns, especially in the third slice where the umbilical cord insertion is far away, which were not visible in conventional IVIM D * maps (Fig. 4h). Figure 5 shows the fetal brain of the same subject reported in Fig. 4 and shows that the twoperfusion model seems to better highlight perfusion differences in tissues than the conventional IVIM model.Indeed, in Fig. 5c the cerebral ventricles membranes clearly have higher f 2 values, whereas they are not visible in the conventional f IVIM map. 25 Moreover, D * 2 (Fig. 5g) has a homogeneous value inside the ventricles.
In Fig. 6, the parametric maps of an FGR and a percreta placenta are displayed.In the DWIs shown in Fig. 6a, the placenta of the FGR subject is outlined in red, whereas the bladder of the percreta patient is outlined in yellow.In the maps of FGR, a placental lacuna is outlined in light blue, where perfusions and diffusions are higher than the surrounding tissues.The conventional IVIM D * (Fig. 6h) does not show the placental lacuna, which is visible in all the other maps.The accretion zone of the percreta placenta is outlined in dark blue and shows high values of f 1 (Fig. 6b), whereas the bladder on the top is characterized by the lowest f 1 values (Fig. 6b).The accretion zone is also characterized by the lowest values of f 2 (Fig. 6c).Even though the accreta placenta is heterogeneous in the f IVIM map (Fig. 6d), the accretion zone has fewer sharp edges compared to the f 1 and f 2 maps (shown in Fig. 6b,c, respectively).The bladder of the accreta placenta is characterized by high values of diffusion given by the urine presence (Fig. 6e), whereas it is characterized by the slowest values of D * 1 compared to the entire placenta, which is highly perfused (Fig. 6f).Finally, the D * 2 parametric map in Fig. 6g shows the lobes' structures of the accreta placenta, which are partially visible in the conventional IVIM D * map displayed in Fig. 6h.

Two-Perfusion and IVIM Metrics
ANOVA showed that both f 2 and f IVIM parameters were significantly higher in normal placentas than those in the FGR placentas (Table 2).Moreover, f 2 and f IVIM were significantly higher on the fetal compared to the maternal side in normal placentas (see Fig. 7a-c and Table 2).
In Fig. 6b,c, the accretion zone has a higher value of the fastest perfusion fraction f 1 than that in the normal placenta.In particular, the entire percreta and increta placentas have the highest f 1 values (see Fig. 7d,f) with a large size effect (Cohen's d = À2.66,Fig. 7f).Moreover, a large effect size was found considering the discriminant power of the f 2 parameter between normal fetal and increta and percreta placenta groups (Cohen's d = 1.12,Fig. 7g).It was also found that f IVIM is higher in the normal fetal side placenta than in the accretion zone with a small effect size (d = 0.26).
Perfusion fraction f 1 was significantly different between the fetal side of SGA and FGR placentas whereas there was no significant difference (P-value = 0.26) in f IVIM (Fig. 7a,c).
Normal placentas had a significant negative correlation between the diffusion coefficient D and the perfusion fractions f IVIM and f 2 (ρ = À0.36 and ρ = À0.38,respectively, in the fetal side and ρ = À0.56 and ρ = À0.50, respectively, in the maternal side) as shown in Fig. 8 (see also Fig. S2 and Tables S4-S6 in Supplemental Material).The accretion zone showed a significant positive correlation between the slowest perfusion fraction f 2 and the GA (ρ = 0.90), whereas the negative correlations between f 1 and GA (ρ = À0.77and P-value = 0.05) and between f IVIM and GA (ρ = À0.63 and P-value = 0.14) did not achieve statistical significance.The normal placenta did not show any correlation between the IVIM perfusion fraction and the gestational age (ρ = À0.31 and P-value = 0.051 in the fetal side and ρ = À0.24 and P-value = 0.12 in the maternal side) or the f 1 and the GA (ρ = À0.24 and P-value = 0.12 in the fetal side and ρ = À0.13 and P-value = 0.40 in the maternal side) or the f 2

Discussion
Since the placental tissue is a complex tissue from a vascular point of view, and most of the placental pathologies are related to vascular dysfunctions, in this work, we have used two-perfusion and IVIM metrics to better describe the complex perfusion in the placenta.
Apart from the number of subjects analyzed and the mean GA of the individual groups, our IVIM analysis of the placenta may differ from those reported in the literature due to the different types of image-denoising treatment.Subsequently, we will discuss the results obtained with the twoperfusion model highlighting the possible advantages of its use, compared to the IVIM model.

Conventional IVIM Model
The IVIM model is currently widely used in placenta MRI studies to estimate perfusion without employing exogenous contrast agents, 13,14,16,[26][27][28][29][30][31][32][33][34] with the perfusion fraction f IVIM being sensitive to changes of perfusion inside the placental tissues.In agreement with previous IVIM studies, f IVIM was significantly higher on the fetal compared to the maternal side reflecting the known physiology of the organ: the placenta's fetal compartment is characterized by the villous trees, so it is  more perfused than the maternal side where the blood diffuses in the intravillous space. 26,27,35Some studies have shown a positive or quadratic correlation between f IVIM and GA in the normal placenta, 14,26,34,36 whereas other studies have found a negative correlation between f IVIM and GA. 28,30n this study, no significant correlation was found between the perfusion fraction and GA in the normal placenta, as in the study of Moore et al. 37 Indeed, Moore et al. 37 suggests that the blood volume, but also the volume of the placenta, increases with GA, so that the perfusion fraction does not change.A possible explanation for the negative correlation between f IVIM and the diffusion coefficient D requires information provided by the parameter f 2 of the two-perfusion model and it will be discussed later, in the two-perfusion subsection.
In agreement with Hutter et al., 29 no significant correlations were found between the diffusion coefficient and the GA in normal placentas.
It has previously been suggested that f IVIM may be a potential marker for placental pathology such as the FGR.4,35 In this current study, the fetal side of normal placentas had slightly (but not significantly) higher values of perfusion fraction compared to those of the accretion zone in pathological placentas, in accordance with Bao et al. 16 who found lower values of the perfusion fraction in the placenta accreta compared to the values in a healthy pregnancy.Considering that in placenta accreta, the trophoblastic invasion extends beyond the normal limit and the placental villi are not contained in the decidual uterine cells, as is normally the case, but extend into the myometrium, the perfusion fraction was expected to be a discriminatory parameter between normal and accreta placenta.

Two-Perfusion IVIM Model
Two-perfusion maps show the potential of the two-perfusion model to highlight particular placental areas, which may be useful for diagnostic purposes or to add information not obtainable from IVIM maps.
In this study, the perfusion fraction f 2 , which is related to the slowest perfusion compartment, was significantly higher on the fetal side of normal placentas than on the maternal side.This result may be interpreted by considering the trophoblastic cells: trophoblasts are responsible for nutrients exchange inside villous trees; thus, they are concentrated on the organ's fetal side.This interpretation is corroborated by the FGR placenta results.In fact, the f 2 parameter is lower in pathological subjects compared to the control group showing a lack of exchanges due to the trophoblastic infiltration on the uterine spiral arteries. 38Moreover, a negative correlation was found between the f 2 parameter and the diffusion coefficient D on the fetal side: the trophoblastic infiltration causes an increase in the blood pressure inside the spiral arteries.This pressure increases the diffusion inside the intravillous space promoting the dispersion of nutrients and decreasing the capability of exchanges between mother and fetus.In contrast to Antonelli et al., 33 no significant differences were found in the IVIM perfusion fraction f IVIM between healthy and SGA subjects.However, the f 1 parameter was significantly different between the fetal side of FGR and SGA subjects.This result may suggest an offsetting effect, whereby the SGA placenta tries to overcome the   difficulty of exchange of nutrients by increasing the fetal fastest perfusion activity of the villous trees.
According to the Cohen's d, the f 1 perfusion fraction discriminates the normal pregnancies and the accreta placenta: the accretion zone is characterized by a higher value of fast perfusion fraction than the healthy subject, especially for percreta placentas where the accretism could involve surrounding organs and could cause hemorrhages during the delivery.The high values of the fastest perfusion fraction f 1 may be due to the different vascular architecture on the accretion zone. 39Conversely, the f 2 perfusion fraction is lower in the case of accretism, showing possible impediments of slow perfusions.Although f 1 and f IVIM correlated negatively with the GA, a positive correlation was found between the f 2 parameter and the GA.A possible explanation for these trends is given by the aging of the placenta.As the placenta ages, it may have less need to increase its vascularity in the anchoring area, which may bring to a decrease in the fastest perfusion activity and an increase in exchange between the infiltrating villi and the maternal blood.
The possible existence of multiple microvascular environments has been also found by Slator et al. 40 : they investigated placental tissues by simultaneously probing the diffusivity and the T2* relaxation time.The T2*-ADC spectra from the inverse Laplace transform of the signal from healthy subjects showed three separate peaks reflecting the three diffusion compartments hypothesized in the two-perfusion model.Moreover, Slator et al. 40 found the absence or reduction of one or two peaks in pathological subjects and lower values of T2* suggesting a deficiency in these compartments.This trend is in accordance with our results since we found that the slowest perfusion fraction f 2 was lower on FGR than in healthy placentas.
A later study by Slator et al. 11 suggested that anisotropic models would better describe placenta physiology.However, these need more gradient directions, which would result in an important increase in the total acquisition time of the experiment.In this work, the quantification of several perfusion and diffusion components in different placenta sites, tries to study the placenta's complex vascularization to be potentially useful for the medical diagnosis of placental impairment.

Limitations
In general, the limitations of this work are related to the limitation of IVIM technique.The most important disadvantage of the IVIM technique is the lack of standardization of the acquisition parameters and the various algorithms used for the quantitative analysis of the images.Furthermore, the sensitivity of the IVIM MRI depends on the number and the distribution of the b-values used.Therefore, due to the lack of standardization of the IVIM technique, significant variance in the calculated parameters was observed between studies, and, to date, no values for normal organs have been well established.In particular, in this work, the number of pathological subjects was low and no measure of inter-observer reproducibility was presented.Nevertheless, results reported here suggest that the two-perfusion model may be useful for studying tissues characterized by two perfusion compartments, one slower, because it is modulated by the passage of fluids through a membrane, and a faster one, in general, associated with perfusion of blood in microcapillaries.

Conclusions
The two-perfusion IVIM model, where f 1 is the fastest perfusion fraction related to perfusion in microcapillaries and villi and f 2 is the slowest perfusion fraction related to the trophoblastic cells' perfusion, may provide complementary information to IVIM parameters that may be useful in identifying placenta impairment.

FIGURE 1 :
FIGURE 1: Flow-chart of the study.
5 T Siemens Avanto (Erlangen, Germany) clinical scanner with an eight-channel body coil without any parallel MRI reconstruction techniques.The acquisition protocol consisted of a diffusion weighted echo-planar imaging spin echo sequence with TR/TE = 3900/74.8msec, bandwidth 1184 Hz/px, matrix size of 192 Â 192, FOV = 220 Â 220 mm 2 , slice thickness of 5 mm, and 18 to 30 slices, depending on the placenta's size.The diffusion encoding gradients were applied along three non-coplanar directions using 10 different b-values (0, 10, 30, 50, 75, 100, 200, 400, 700, and 1000 sec/mm 2 ), and the signal over the three directions was averaged.The number of averaged signals was NS = 4 for each b-value, thus the total duration of the protocol was $15 minutes.

FIGURE 2 :
FIGURE 2: Denoising of DWI.The original DWI of a normal placenta at b = 50 sec/mm 2 is shown, (a); the same slice following noise correction (b).A plot of SNR vs.(b) is shown in (c).

FIGURE 3 :
FIGURE 3: Example of a fit to DW-data obtained in the fetal ROI.The error bars were evaluated propagating the uncertainty on the noise c σ 2 L on the voxels inside the ROI.The Akaike weights are w mono AIC ð Þ¼6:2e À 06, w IVIM AIC ð Þ¼5:9e À 2, w 2perf AIC ð Þ¼0:94.

FIGURE 4 :
FIGURE 4: (a) The DWI section shows three different slices of the same normal placenta (GA = 22.4 weeks): the first (upper) slice shows the fetal brain, the umbilical cord (outlined in red) and a section of the placenta divided into maternal (blue) and fetal (green) sides.The second slice focuses on the umbilical cord insertion (in red), and the third slice is an overview of the entire central placenta (in red the umbilical cord).(b) f 1 parametric maps: the maternal decidua shows the highest values maybe due to the spiral arteries insertion; (c) f 2 parametric maps.(d) f IVIM parametric maps; (e) The diffusion D maps shows high values of diffusion in the amniotic liquid as it is free water like.(f) D1* parametric maps; (g) D2* maps show interesting patterns that can be interpretated as the cotyledon structures in the placental surface.(h) D*IVIM maps.

FIGURE 5 :
FIGURE 5: (a) Zoomed images of the fetal brain visible in the upper placenta slice of Fig. 4. (b) The f 1 map shows lower values in fetal brain compared to those in ventricular space.(c) The perfusion fraction f 2 is higher around ventricles' space, showing a probable exchange between ventricles and brain's tissues through the ventricular membrane.(d) f IVIM map; (e) the diffusion coefficient D map.(f) D1* perfusion coefficient map; (g) The D2* perfusion coefficient map shows perfusional activity inside the ventricles; whereas, it shows lowest values in the cerebral tissue.(h) D* IVIM perfusion coefficient.

FIGURE 6 :
FIGURE 6: (a) DWI of an FGR placenta (GA = 19.7 weeks, outlined in red) and a percreta placenta with bladder infiltration (GA = 28.6 weeks, the bladder is outlined in yellow).(b) f 1 parametric maps: the FGR placenta shows a placental lacuna (outlined in light blue) where perfusion fraction is higher due to a placental lacuna; the accretion zone on the placenta accreta is outlined in dark blue and shows high values of f 1 .(c) f 2 parametric maps: the FGR placenta has the highest values on the fetal side; the accretion zone (dark blue) is characterized by the slowest values of f 2 .(d) f IVIM parameter maps.(e) Diffusion D parameter maps: the placental lacuna of the FGR subject is clearly visible and this pattern could be due to the trophoblastic invasion that increases the diffusivity inside the maternal side of FGR subjects causing placental lacunae.(f) Perfusion coefficient D1* maps.(g) D2* parametric maps show the lobes' structures of the accreta subject.(h) IVIM D* maps.

FIGURE 7 :
FIGURE 7: Boxplots of the perfusion fractions for the fetal and maternal placenta ROIs: (a) f 1 parameter; (b) f 2 parameter; and (c)f IVIM .The group with placenta accretism was highly heterogeneous as shown in the plot of f 2 vs. f 1 (d).The percreta and increta placenta group had the highest f 1 values (d) and a large size effect (Cohen's d = À2.66)(f).A large effect size was also found for the f 2 parameter between the normal fetal and increta and percreta placenta groups (Cohen's d = 1.12) (g).(e) f IVIM has a small effect size between normal and increta and percreta groups (Cohen's d = 0.32).

TABLE 1 .
Pregnancy Women Cohorts

TABLE 2 .
ANOVA with Dunn-Sidak Post Hoc Correction and Cohen's d Values Note: Statistical significant p-values are in bold.FGR: fetal growth restriction; SGA: small for gestational age.