Research ArticleImaging

Noninvasive high-resolution electromyometrial imaging of uterine contractions in a translational sheep model

See allHide authors and affiliations

Science Translational Medicine  13 Mar 2019:
Vol. 11, Issue 483, eaau1428
DOI: 10.1126/scitranslmed.aau1428

Reconstructing contractions

Monitoring uterine contractions during labor is critical to ensure good health of the mother and of the baby. Current methods present several limitations, including low resolution and invasiveness. Now, Wu et al. developed a noninvasive three-dimensional electromyometrial imaging called EMMI, able to monitor uterine contractions with high spatial and temporal resolution. Combining data obtained from electrodes placed on the abdomen with magnetic resonance imaging, EMMI reconstructed uterine contraction patterns in sheep. The results matched reconstructions obtained with invasive electrodes placed on the uterine surface. EMMI might provide an easily accessible method to help doctors during labor and to better understand uterine electrophysiology and pathophysiology.

Abstract

In current clinical practice, uterine contractions are monitored via a tocodynamometer or an intrauterine pressure catheter, both of which provide crude information about contractions. Although electrohysterography/electromyography can measure uterine electrical activity, this method lacks spatial specificity and thus cannot accurately measure the exact location of electrical initiation and location-specific propagation patterns of uterine contractions. To comprehensively evaluate three-dimensional uterine electrical activation patterns, we describe here the development of electromyometrial imaging (EMMI) to display the three-dimensional uterine contractions at high spatial and temporal resolution. EMMI combines detailed body surface electrical recording with body-uterus geometry derived from magnetic resonance images. We used a sheep model to show that EMMI can reconstruct uterine electrical activation patterns from electrodes placed on the abdomen. These patterns closely match those measured with electrodes placed directly on the uterine surface. In addition, modeling experiments showed that EMMI reconstructions are minimally affected by noise and geometrical deformation. Last, we show that EMMI can be used to noninvasively measure uterine contractions in sheep in the same setup as would be used in humans. Our results indicate that EMMI can noninvasively, safely, accurately, robustly, and feasibly image three-dimensional uterine electrical activation during contractions in sheep and suggest that similar results might be obtained in clinical setting.

INTRODUCTION

Each year, about half a million women deliver preterm in the United States (1), which puts their babies at increased risk of mortality and long-term neurological morbidity (2, 3). Although 45% of preterm births begin with spontaneous preterm labor (4), about half of women who go into preterm labor go on to deliver at term (5). The differences between preterm contractions that cease and those that do not are largely unknown. Moreover, we have limited understanding of normal term labor, which prevents us from answering basic questions such as where electrical activity starts during a contraction, how electrical activity propagates to yield localized or global contractions, and whether contractions always start in the same location and propagate in the same manner.

Several techniques have been developed to monitor uterine contractions; however, they present several limitations. Clinicians can manually palpate a patient’s abdomen during labor, but this method is time consuming and observer dependent. In alternative, intrauterine pressure (IUP) measurement can be performed by transvaginally placing an IUP catheter into the uterus. Although IUP measurement is commonly used and is considered the gold standard for monitoring contractions, this method requires an invasive procedure that poses risks (such as infection) to the mother and potentially to the neonate (6) and thus is only performed when medically necessary. A tocodynamometer (TOCO) transducer can be placed on a patient’s abdomen to measure small contour displacements caused by uterine contractions. TOCO transducers are easy to apply and provide information such as contraction frequency and length, but the resulting data only weakly correlate with contraction amplitude data obtained by IUP measurement (7). In addition, TOCO requires frequent transducer adjustment and is prone to artifacts caused by other maternal and fetal movements. Magnetomyography (MMG) can be performed by placing an array of sensors in a fix-contoured hemisphere to cover the front of the abdomen during early labor (8, 9). Although MMG data correlate with contractile events perceived by mothers and provide distribution maps of local uterine activity, this method does not provide a three-dimensional (3D) view of the entire uterus and requires a large piece of specialized equipment in a magnetically shielded room. Electrohysterography (EHG) uses a few electrodes placed on the patient’s abdomen to measure changes in electrical potential. EHG is promising in that it can detect human uterine contractions (1012) and can correlate electrical signal properties with preterm labor (1315). Recently, EHG spatial coverage and resolution were improved by the use of 4 by 4 array sensors (covering about 8 cm by 8 cm) (16, 17) and 8 by 8 grid sensors (covering 2.8 cm by 2.8 cm) (18). Although this work has shed light on electrical signal propagation, EHG is limited to measuring a small area on the maternal abdomen. Last, uterine electromyography (EMG) can be performed by placing electrodes directly on the uterine surface. EMG has been used in animal studies both in vivo (19) and in vitro (20, 21), but the invasiveness of the procedure prevents its use in humans.

To overcome the limitations of current technologies, we adapted the principles of electrocardiographic imaging (ECGI) (2227). ECGI combines patient-specific body-heart geometry (obtained by computed tomography) and detailed body surface electrical maps (obtained from up to 256 electrodes embedded in a vest) to noninvasively image electrical activation, propagation, and reentry circuits of the heart at high temporal and spatial resolution with remarkable accuracy (2428). On the basis of the similar idea, we developed an electromyometrial imaging (EMMI) method. In EMMI system, we used magnetic resonance imaging (MRI), which is safe in pregnancy (2931), to create a body-uterus geometry. Next, we recorded electrical activity data from up to 256 electrodes placed on the abdomen and then used EMMI software to map the electrical signals onto the uterus geometry. We could thus noninvasively monitor the initiation and propagation of uterine contractions by tracking electrical signals across the entire uterine surface.

Here, we used a term pregnant sheep model to show that EMMI provided accurate measurements, as uterine surface electrical potential maps reconstructed from body surface potentials matched those measured from electrodes placed directly on the uterus during electrical pacing. Next, we showed that EMMI could robustly reconstruct uterine electrical activation maps during oxytocin-induced contractions, even in the presence of simulated Gaussian noise and geometric deformations. Last, we showed that EMMI is feasible because we could use it to noninvasively map induced uterine contractions. Given its safety, accuracy, robustness, and feasibility, EMMI may become a resource for evaluation of uterine contractility in humans.

RESULTS

Overview of EMMI

To create a 3D subject-specific body-uterus geometry, MRI was performed after applying up to 256 MRI markers to the subject’s body surface around the abdomen and lower back. When the subject was in labor, the MRI markers were replaced with body surface electrodes in the same locations, and multichannel body surface potentials were simultaneously recorded with a portable electrical mapping device. Last, as was done in ECGI (23, 32), EMMI software used the method of fundamental solution (MFS) to solve Laplacian partial differential equations and combine electrical signals with the uterus geometry to generate uterine surface potential maps (electrical activity across the uterus at a single time point; Fig. 1A). These maps were used to derive electrograms (electrical waveforms over time at each uterine site) and isochrone maps (time of activation at each point across the entire uterine surface).

Fig. 1 EMMI system and method to assess accuracy.

(A) MRI scans were acquired and then segmented to generate body-uterus geometry. On the body surface, up to 256 electrodes were placed in the locations of the corresponding MRI markers. Body surface electrograms were recorded and mapped onto body surface potentials. The measured body surface potentials and the body-uterus geometry were combined by EMMI software to generate reconstructed uterine surface potentials (spatial potential distribution on the 3D uterine surface at each instance in time). (B) The sheep uterus was surgically exposed, and an elastic sock containing 64 electrodes was slipped over the uterus. An electrical pacing lead (indicated by an asterisk) was placed onto the uterine surface. After closing the abdomen, body surface electrodes were placed in their original locations. When pacing the uterus through the pacing lead, the uterine and body surface electrograms were recorded simultaneously. The uterine surface electrograms were directly mapped onto MRI-derived uterine surface to generate measured uterine surface potentials. Last, the EMMI-reconstructed uterine surface potentials (A) were qualitatively and quantitatively compared to the measured (B) uterine surface potentials.

EMMI accuracy

To assess EMMI accuracy, we used a scheme similar to that used to validate ECGI, in which the epicardial potentials measured from an animal heart were compared to those reconstructed from body surface potentials measured from a human torso–shaped tank in which the animal heart was suspended (23, 33). We performed MRI on the anesthetized sheep wearing MRI markers around the lower abdomen and back (Fig. 1A) to generate a body-uterus geometry. We then surgically exposed the uterus, slipped an elastic sock containing 64 electrodes over the uterus, placed a cardiac pacing lead directly on the uterine surface through the sock, and returned the uterus to its original location. After closing the abdomen, 192 electrodes were placed on the body surface in the same position as the MRI markers. While we paced the uterus with a well-controlled electric pulse from the pacing lead, we simultaneously recorded electrical potentials from body surface electrodes and uterine surface electrodes. We then used EMMI software to reconstruct uterine surface potential maps from the body surface potentials and MRI-derived body-uterus geometry. Last, we compared the reconstructed uterine surface potential maps with those measured directly from the uterine surface during the pacing episodes (Fig. 1, A and B).

We recorded a total of 118 independent pacing pulses from two sheep. The pacing lead (indicated by an asterisk in Fig. 2) was placed in the top segment of the uterus near the fundus in sheep A (Fig. 2A) and in the middle segment of the uterus in sheep B (Fig. 2B). Figure 2 shows measured and EMMI-reconstructed uterine surface potential maps in a right lateral view during three independent pacing pulses for sheep A (Fig. 2A) and B (Fig. 2B) (see numerical data in data file S1). In these maps, each generated at a specific point in time, warm colors denote positive potentials, and cool colors denote negative potentials. The measured potential maps only cover part of the uterine surface because of the limited number of sock electrodes. In contrast, the EMMI-reconstructed uterine surface potential maps (generated from the body surface potentials) represent the potential distribution pattern of the entire uterus. The mean electrode spacing in the measured potential maps was 47 ± 21 (SD) mm for sheep A and 48 ± 18 mm for sheep B. The mean reconstruction point spacing in the EMMI-reconstructed uterine surface potential maps was 31 ± 8 mm for sheep A and 32 ± 7 mm for sheep B (see detailed analysis in data file S2). The null hypothesis that the spatial resolution of EMMI reconstruction is same with measured spatial resolution was rejected by Wilcoxon rank sum test with P < 2.2 × 10−16. We also noted that, in the measured potential maps, there were interpolation (triangle-shaped) artifacts due to limited number of sock electrodes.

Fig. 2 Right lateral view of the uterine surface potential maps during pacing pulses.

Potential maps at the peaks of three pacing pulses for sheep A (A) and sheep B (B). Electrograms at top were from sites of the pacing leads (white asterisks). N and N′ denote negative potential centers in the measured and EMMI-reconstructed potential maps, respectively, whereas P and P′ denote positive potential centers in the measured and EMMI-reconstructed potential maps, respectively. Dashed lines denote vectors connecting the negative and positive centers. Spatial correlation coefficients (CCs) of the potential maps shown in this figure (defined in Eq. 7) were computed at the peak of each pacing pulse. Distance error of negative and positive potential centers between the measured and EMMI-reconstructed potential maps and CC of potential maps were analyzed during all pacing pulses for sheep A (n = 138) and sheep B (n = 390), as summarized in Table 1, and detailed data are provided in data file S3.

In both the directly measured and the EMMI-reconstructed uterine surface potential maps, we observed negative potential centers (labeled as N and N′ in Fig. 2) adjacent to the pacing leads (white asterisk in Fig. 2). In addition, we observed nearby positive potential centers (labeled as P and P′ in Fig. 2). These potential centers were in similar locations in the measured and EMMI-reconstructed potential maps. Specifically, in sheep A (Fig. 2A), the negative and positive potential centers differed by 16.5 ± 7.2 mm (means ± SD) and 28.8 ± 11.4 mm, respectively, between the measured and EMMI-reconstructed potential maps, and the angles of the vectors connecting the negative and positive centers differed by 6.1 ± 6.5° (n = 138, from 30 pacing pulses; see data file S3). Similarly, in sheep B (Fig. 2B), the negative and positive potential centers differed by 2.9 ± 0.0 mm and 8.1 ± 6.3 mm, respectively, between the measured and EMMI-reconstructed potential maps, and the angles of the vectors connecting two centers differed by 7.5 ± 8.2° (n = 390, from 78 pacing pulses; see data file S3).

To quantitatively assess EMMI reconstruction accuracy, we calculated CCs (see Eq. 7 in Materials and Methods) between measured and EMMI-reconstructed uterine surface potential maps. In this case, CC values reflect the correlation between measured and EMMI-reconstructed potentials at a single time point for 62 electrode sites in sheep A and 49 electrode sites in sheep B (64 electrodes were placed on the uterus in both sheep, but data from 2 electrodes in sheep A and 15 electrodes in sheep B were discarded owing to poor electrical contact with the uterus). CC is bounded between 0 and 1, with high similarity closer to 1 and low similarity closer to 0. Potential map CC values had a median of 0.71 [first quartile (Q1) = 0.67, Q3 = 0.74] for sheep A (n = 138; see data file S3) and 0.83 (Q1 = 0.81, Q3 = 0.84) for sheep B (n = 390; see data file S3). Movies S1 and S2 include potential map movies for all frames during each pacing shown in Fig. 2. CC values suggest that EMMI could accurately reconstruct uterine surface potential maps from body surface potentials during uterine pacing in sheep.

EMMI robustness

We wanted to determine how robustly EMMI could measure uterine electrical activation in circumstances that would be experienced clinically. Specifically, could EMMI measure oxytocin-induced contractions, and would the EMMI reconstruction be negatively affected by maternal or fetal movement that is likely to occur after the MRI but before or during the electrical recording? In addition, would the EMMI reconstruction be affected by electrical noise in the recording room, such as that produced by nearby equipment? To answer these questions, we used a scheme (Fig. 3), similar to a well-established scheme used to validate ECGI (34, 35), to evaluate the robustness of EMMI electrogram, potential map, and isochrone maps.

Fig. 3 Method to assess EMMI robustness.

MRI scans were acquired to provide body-uterus geometry. Next, the uterus was surgically exposed, and an elastic sock containing up to 128 electrodes was slipped over the uterus. The uterine surface potentials were recorded at 2048-Hz sampling rate. Next, adding noise, deformation, or both combined in bioelectric field computation, the body-uterus geometry and experimentally measured uterine surface potentials were used to generate the body surface potentials. EMMI software was then used to reconstruct uterine surface potentials. Last, the EMMI-reconstructed and measured uterine surface potentials were compared.

EMMI electrogram

We first evaluated electrograms in an individual “episode,” defined as a recording segment of 300 ± 70 s, which typically contained five to nine electrical bursts at each contracting site. To assess accuracy, we calculated CC, which reflects the correlation between measured and EMMI-reconstructed electrograms over time at each site on the uterine surface, and relative error (RE; see Eq. 8 in Materials and Methods), which reflects the difference in magnitude between measured and EMMI-reconstructed electrograms. Zero RE suggests that there is no difference, and high RE suggests a large difference. Therefore, RE has no upper bound. For episode 1, we used CC and RE to compare the measured uterine electrograms at all sites to the corresponding EMMI-reconstructed uterine electrograms to which we added noise, geometrical deformation, or both noise and deformation. In the presence of noise, the median CC was 0.88 (0.73, 0.96) and the median RE was 0.55 (0.32, 0.78); in the presence of deformation, the median CC was 0.86 (0.71, 0.94) and the median RE was 0.62 (0.41, 0.83); and in the presence of both noise and deformation, the median CC was 0.86 (0.70, 0.94) and the median RE was 0.63 (0.41, 0.85) (Fig. 4A and numerical data in data file S4). We next compared the measured and EMMI-reconstructed uterine electrograms at five representative locations (sites 15, 19, 21, 48, and 53; Fig. 4B). At these locations, the EMMI-reconstructed electrical burst morphologies were close to the measured electrical burst morphologies (CC from 0.70 to 0.95) under all three conditions (Fig. 4C). The reconstructed electrical amplitudes were also well preserved at all uterine locations (RE from 0.34 to 0.71). The measured and EMMI-reconstructed uterine surface electrograms at two representative locations in episodes 2 to 8 are included in fig. S1.

Fig. 4 Evaluation of EMMI-reconstructed uterine surface electrograms in episode 1.

(A) Box plot of CCs (blue, diamond checkered pattern; defined in Eq. 7) and REs (orange, diagonal pattern; defined in Eq. 8) comparing EMMI-reconstructed with measured uterine surface electrograms under the indicated conditions (n = 52; see data file S4). Solid horizontal lines indicate Q1, median, and Q3. (B) Left and right lateral view of sheep uterus. The numbers differentiated by various shapes indicate the discrete uterine surface sites where measured and reconstructed uterine surface electrograms are compared in (C). (C) Measured and EMMI-reconstructed electrograms (0 to 200 s) from the indicated sites. Analysis of all episodes is presented in Table 1 (n = 595) and data file S5.

We next evaluated the accuracy of the EMMI-reconstructed electrograms from all eight episodes. Across all eight episodes, CCs were at 0.85 (0.72, 0.95), 0.83 (0.69, 0.93), and 0.83 (0.68, 0.93) under noise, deformation, and both noise and deformation, respectively; whereas, REs were at 0.55 (0.36, 0.78), 0.62 (0.42, 0.86), and 0.63 (0.43, 0.87) under noise, deformation, and both noise and deformation, respectively (n = 595; see Table 1 and detailed analysis in data file S5). Together, these data indicate that EMMI was able to consistently reconstruct accurate uterine electrograms from body surface potential data even in the presence of added noise, geometrical deformation, and both noise and deformation.

Table 1 Accuracy analysis of all pacing data, electrograms, potential maps, and isochrone maps reconstructed by EMMI.
View this table:

EMMI potential maps

We next compared potential maps measured at the uterine surface to those reconstructed by EMMI in the presence of noise, geometrical deformation, and both noise and deformation. Figure 5 shows three representative potential maps at different time points (denoted by red arrows in the electrograms) during episode 1 (see numerical data in data file S6). The potential distribution patterns reconstructed by EMMI under noise, geometrical deformation, and both noise and deformation were similar to those directly measured on the uterine surface. In Fig. 5A, the potential distribution patterns labeled A1, A2, and A3 in the measured potential maps were persevered in EMMI-reconstructed potential maps under all three conditions. Quantitatively, as labeled in Fig. 5, the spatial CC values for the entire potential maps were high (0.87 to 0.93), and the spatial RE values were low (0.38 to 0.49). We analyzed CC [0.80 (0.71, 0.86), 0.78 (0.68, 0.85), and 0.77 (0.67, 0.84)] and RE [0.59 (0.48, 0.73), 0.64 (0.52, 0.78), and 0.64 (0.53, 0.78)] under noise, deformation, and both noise and deformation, respectively, for 28,120 potential maps from eight episodes, as summarized in Table 1 (see detailed analysis in data file S7). The overall similarity between the measured and EMMI-reconstructed potential maps indicates that EMMI can accurately reconstruct uterine surface potential patterns during uterine contraction even in the presence of additional noise, geometrical deformation, and both additional noise and deformation.

Fig. 5 Evaluation of EMMI-reconstructed uterine surface potential maps during episode 1.

Measured potential maps and reconstructed maps with noise, deformation, or both noise and deformation shown at time instances (red arrows in electrograms) of 0 s (A), 11 s (B), and 20 s (C). The electrograms shown were from sites A1, B1, and C1. CCs (defined in Eq. 7) and REs (defined in Eq. 8) shown in this figure were computed at their corresponding time instance. Analysis of all potential maps during contraction is presented in Table 1 (n = 28,120) and data file S7.

Isochrone maps

For assessment of the robustness of EMMI reconstruction, we generated isochrone maps to reflect the electrical activation pattern of the uterus during a particular observation window (the observation window is defined in Materials and Methods). We constructed isochrone maps by using a heat map to denote the activation time of each uterine site; warm colors indicate uterine regions that activated early, and cool colors indicate regions that activated late. Figure 6 shows isochrone maps for two observation windows, one from 0 to 59 s and the other from 173 to 227 s in episode 1 (see numerical data in data file S8). In the first observation window (Fig. 6A), we observed early activation (red) in a large region along the fetal spine (maternal ventral) and a small region at left lateral and maternal dorsal, which then locally propagated to nearby regions (yellow and then green in color). In the second observation window (Fig. 6B), we observed early activation in three connected regions at right lateral, which then propagated to left lateral. The uterine sites marked by the black dashed circles and squares activated early in both observation windows. Movies S3 to S5 include activation movies for the two observation windows. EMMI-reconstructed isochrone maps in the presence of noise, deformation, or both preserved activation patterns in directly measured isochrone maps during both observation windows (observation window A: CC = 0.96, 0.95, and 0.95 and RE = 0.14, 0.14, and 0.14 under three conditions, respectively; observation window B: CC = 0.99, 0.99, and 0.99 and RE = 0.01, 0.01, and 0.01 under three conditions, respectively). For all 25 isochrone maps from eight episodes, CCs are 0.99 (0.96, 1.00), 0.98 (0.95, 0.99), and 0.97 (0.94, 1.00), whereas REs are 0.01 (0.00, 0.06), 0.02 (0.01, 0.08), and 0.03 (0.01, 0.08) under noise, deformation, and both noise and deformation, respectively (Table 1 and detailed analysis in data file S9). These results suggest that EMMI can accurately reconstruct isochrone maps in the presence of noise, deformation, or both.

Fig. 6 Evaluation of EMMI-reconstructed activation isochrone maps in episode 1.

Measured and EMMI-reconstructed activation isochrone maps with noise, deformation, or both noise and deformation are shown during (A) observation window A (0 to 59 s) and (B) observation window B (173 to 227 s). The electrogram represents the site marked with an asterisk. In the isochrone maps, red indicates the earliest activation, blue indicates the latest activation, and the darkest blue, labeled “inf,” denotes regions in which no activation was recorded during the observation window. Black dashed circles and squares denote uterine surface areas that activated early in both windows. CCs and REs shown in this figure were computed at their corresponding observation windows. Analysis of all isochrone maps is presented in Table 1 (n = 25) and data file S9.

EMMI feasibility

Last, we evaluated the feasibility of EMMI being used to noninvasively map uterine surface potentials from measured body surface potentials. Thus, we analyzed data collected from the body surface of four sheep after at least two oxytocin boluses were delivered before we performed surgery. Body surface electrical activity bursts during uterine contractions were measured and confirmed by two obstetricians (A.G.C. and J. Rhoades) and one veterinary surgeon (M.T.). The three contiguous uterine contractions were reconstructed by EMMI using the measured body surface potentials and MRI-derived body-uterus geometry. The detailed activation sequences were shown in Fig. 7 and data file S10. Other nine uterine contractions were reconstructed by EMMI, and the isochrone sequences were shown in fig. S2 and data file S11. The results indicated that EMMI is feasible to noninvasively image uterine contraction.

Fig. 7 EMMI-reconstructed activation isochrone maps of oxytocin-induced contractions.

Three contiguous contractions—0 to 22 s (A), 26 to 50 s (B), and 56 to 84 s (C)—mapped by EMMI. The EMMI-reconstructed electrogram is from the uterine surface site denoted by an asterisk. In the isochrone maps, light pink indicates the earliest activation, blue indicates the latest activation, and the darkest blue, labeled “inf,” denotes regions in which no activation was recorded during the observation window. Other nine uterine contractions were reconstructed by EMMI, and the activation sequences (isochrones) were shown in fig. S2 and data file S11 (total n = 12).

DISCUSSION

Our data provide strong evidence that EMMI is a suitable tool to safely and noninvasively image electrical activity of the uterus during contractions. We showed that potential maps produced by EMMI from recordings made at the body surface were similar to those made by placing electrodes directly on the uterine surface during electrical pacing. In addition, in modeling experiments, we showed that EMMI reconstructions are minimally affected by the realities of clinical scenarios, such as maternal and fetal movement that would occur between the time of the MRI and the electrical recording, and electrical noise emanating from nearby equipment. Last, we showed that EMMI could be used on sheep in the manner we envision it used in humans by deriving body-uterus geometry from MRI, followed by noninvasive electrical imaging. Together, these data indicate that EMMI can combine MRI and body surface electrical measurements to accurately, robustly, and feasibly reconstruct uterine electrical activity during contractions in sheep.

Several reasons suggest that EMMI should be safe for use in pregnant women. MRI has been used to evaluate obstetrical, placental, and fetal abnormalities in pregnant patients during their second and third trimesters (29, 36). There has been some concern on exposure to static and gradient magnetic fields, radio frequency, heating of sensitive tissues (37), and acoustic environment (38); however, in a population-based cohort study involving more than 1.4 million pregnant women, first-trimester MRI showed no association with stillbirth or neonatal death, congenital anomaly, neoplasm, or hearing loss (31). The electrical recording should also be safe because the patient is protected by battery-powered recording box and optical isolation between the recording box and the computer. In addition, the leak current specified by BioSemi is less than 1 μA, well below the safety level of 10 μA (39, 40). Last, the system will automatically shut down when leakage current is above the safety level (40).

EMMI provides several advantages over direct electrical measurement from the uterine surface as is done in EMG. EMMI is noninvasive, whereas EMG requires surgery to expose the uterus and thus is only feasible in animal studies. Moreover, EMMI can image uterine electrophysiology at higher spatial resolution than EMG. This is because the total number of electrodes applied on the uterine surface for EMG is constrained by the complexity of surgery. For example, in our pacing study, we used a sock that contained 64 electrodes. In contrast, by using EMMI, we were able to assess about 180 (or more) reconstruction points (see mesh point index in data file S1). Thus, the mean distance between reconstruction points was substantially smaller than the distance between sock electrodes. This superior spatial resolution translated to more physiologically plausible electrical potential distributions. Whereas potential maps generated from electrodes directly on the uterine surface had interpolation artifacts, those generated by EMMI were ellipsoid, which reflects the anisotropic electrical propagation and is consistent with the shape reported in a study in which electrodes were placed directly on isolated rat uterine tissue (41). Last, we could not completely cover the uterus with the sock and thus could not image the entire uterus by direct measurement. Conversely, EMMI permitted electrical mapping of the entire 3D uterus, thus providing a complete view of electrical activation patterns during uterine contractions.

We noted a few potential limitations in this first EMMI study. First, as mentioned above, we were only able to cover part of the uterus with a sock that contained a low density of electrodes, preventing us from obtaining detailed electrophysiological information about the sheep’s uterine contractions. Second, our EMMI computation assumed that the volume conductor between the body surface and uterus was homogenous without any electrical sources. Under this assumption, we found that activation time and potential patterns were similar between directly measured and EMMI-reconstructed data. Similarly, previous ECGI studies found that ignoring inhomogeneity minimally affected the reconstructed signal morphology (42, 43). However, the inhomogeneity introduced by bone, fat, and skeletal muscle could potentially reduce EMMI’s reconstruction accuracy on the magnitude of the potential. Future work will be aimed at modeling the effects of inhomogeneity to improve EMMI reconstruction. Third, we only assessed electrical burst propagation and did not assess electrical spike propagation (44). Future work will be needed to examine the ability of EMMI to reconstruct data with sufficiently high temporal and spatial resolution to reconstruct single spike propagation. Fourth, to stabilize the inverse solution in EMMI, we used the conventional Tikhonov regularization technique with a regularization parameter determined using composite residual and smoothing operator (CRESO). Although this method is effective and accurate in analyzing cardiac signals in ECGI (23, 32), these parameters may not be optimal for analyzing uterine signals in EMMI because of the differences in signal morphology and amplitude between cardiac and uterine measurements. Future work will be needed to refine the regularization technique and parameters optimal for EMMI. Last, our analysis of EMMI robustness was limited to examining the effects of Gaussian noise and geometrical deformation. We plan to further evaluate the robustness of EMMI in the context of different types and severity of noise and geometrical deformation.

Further development of EMMI will focus on enhancing its value as both a research tool and a clinical tool. As a research tool, EMMI will continue to use MRI to provide accurate geometrical information; as a clinical tool, EMMI needs to adopt portable imaging modalities such as ultrasound to replace MRI in the delivery room and incorporate a simplified electrical mapping system with smaller number of electrodes to effectively map patients during labor. Another important issue will be to determine the utility of EMMI in measuring the propagation velocity of single electrical spike in the myometrium. Several studies reported that propagation velocity correlates with up-regulation of the gap junction protein connexin-43, increased intercellular coupling, and uterine synchronization and maturation (13, 4547). Thus, propagation velocity might be used as a marker of labor progression (13, 18). Regional propagation velocity has been invasively measured and calculated on the uterine surface of rats and guinea pigs (20, 21), but, to the best of our knowledge, no reports have described global propagation velocity of spikes covering the entire uterus to reveal the complete electrical activation pattern. In humans, researchers have attempted to measure propagation velocity by using a small electrode array to measure electrical signals from an area of 2.8 cm by 2.8 cm or 8 cm by 8 cm on the body surface (18, 48, 49). However, the propagation velocity measured from the maternal body surface may be inaccurate, especially when the uterine surface electrical wavefront propagation is not parallel to the axis of a measuring electrode pair on the body surface (44). Given that EMMI can accurately reconstruct isochrone maps over the entire uterus from body surface electrical recordings, we consider the possibility of using EMMI to measure propagation velocity.

Another important aspect of labor that we aim to investigate with EMMI is the characterization of “pacemakers” that initiate contractions in a coordinated fashion during active labor (5052) to generate sufficient IUP to expel the fetus. Similar to the sinoatrial node of the heart, uterus has been thought to have a single pacemaker (53). However, such uterine pacemaker has not yet been found after decades of research (44, 54). Alternative theories have suggested that, instead of having one unique pacemaker as in the heart, the uterus may have multiple pacemakers throughout the uterus (51, 52, 5557). One of the plausible working models of the multiple pacemaker theory suggested that each pacemaker paces a region of myometrium in the uterus, leading to gradually increased IUP, which in turn initiates other pacemaker activities at different locations (55, 58). Consistent with this theory, our sheep EMMI data suggest that early activation occurred at multiple sites on the uterine surface within one observation window. MMG measures the magnetic signals generated by uterine tissue contractions using an array of superconducting quantum interference devices (SQUIDs) on the anterior abdomen surface (8, 9, 59). SQUID studies also provided important supporting evidence for multiple pacemakers theory (59). To better understand the heterogeneous clinical presentation of human uterine contractions, we will focus on determining the number and location of pacemakers on the uterine surface and characterizing their association with labor progression in human EMMI studies.

Next steps for EMMI development include careful testing in the clinical setting to gain a more complete understanding of the electrophysiologic properties involved in term and preterm labor. It is possible that advancing our understanding of labor with EMMI could affect labor management in several ways. First, EMMI could be useful for triaging pregnant women presenting with contractions, on the basis of the imaged pattern of contraction initiation and propagation. Second, EMMI may help assess the effects of medical tocolytic therapy in ways beyond the current TOCO technology. Last, improved understanding of the changes in human uterine electrophysiology and pathophysiology may enable new developments in managing labor that mirror those used to manage cardiac arrhythmias. Possibilities include modulation of sodium channels, pacing strategies to induce or augment labor, and even catheter-based ablation of hyperactive uterine pacemaker regions with a goal to prevent preterm labor.

MATERIALS AND METHODS

Study design

This study was designed to develop EMMI and validate its capability to noninvasively image the 3D electrical activation patterns during uterine contractions. To achieve this goal, we first used pregnant sheep model to simultaneously record the electrical potentials from body surface electrodes and uterine surface electrodes. We validated EMMI’s accuracy by comparing the EMMI-reconstructed uterine potential maps to those measured directly from the uterine surface during the pacing episodes. We next investigated EMMI’s robustness of reconstructing uterine electrical activation in circumstances that would be experienced clinically. Specifically, we evaluate EMMI-reconstructed uterine electrical activation maps during oxytocin-induced contractions, even in the presence of simulated Gaussian noise and geometry deformations. Last, we evaluate EMMI’s feasibility because we can use it to noninvasively map uterine contractions in human. All animal studies and procedures were approved by the Washington University in St. Louis Institutional Animal Care and Use Committee.

Sheep model

To test EMMI, we used Katahdin sheep (19, 60, 61), which are a good model for human pregnancy because they have a similar abdomen size as humans. In addition, properly timed steroid injections can be used to induce sheep labor (61, 62). Although sheep have a bipartite uterus, a single newborn offspring is of similar weight to a human baby (60). We used nine near-term pregnant Katahdin sheep (which have less lanolin skin secretions than other breeds); the first three sheep were used as pilot animals to develop MRI sequences and experimental protocols, and the remaining six sheep were used for electrical recordings. We excluded data from one sheep that had twins and used the remaining five sheep to assess the accuracy, robustness, and feasibility of EMMI.

Sheep preparation

Katahdin sheep (Francis Sheep Farm, an institutionally approved vendor) were obtained between gestation days 140 and 145 (on average, this breed delivers lambs around day 147) of their first or second pregnancies. All procedures were approved by the Washington University in St. Louis Institutional Animal Care and Use Committee. Dexamethasone [16 mg (62, 63)] was intramuscularly administered 24 to 48 hours before the MRI to sensitize the uterine response to oxytocin. Sheep were fasted before anesthesia and on the day of the study. Sheep were anesthetized with ketamine (10 mg/kg) by intramuscular injection and animals were maintained under general anesthesia using inhalational isoflurane (1–4%) during the MRI and surgical procedure. Sheep were shaved free of hair circumferentially from the midthorax to the level of the pelvis.

MRI scan

MRI markers were applied externally around the lower abdomen and back of the sheep. MRI (without any contrast agent) was performed on a 3T Siemens Prisma using a radial volume interpolated breath-hold examination fast T1-weighted sequence with a spatial resolution of 1 mm by 1 mm by 3 mm. Localizer was used to adjust the field of view to cover the entire sheep uterus and cervix. Then, sheep were scanned along the axial direction for about 135 to 200 slices, depending on the size of the sheep.

Surgery and electrical recording

Within 1 hour after the MRI, the sheep was brought to the operating room and placed in left lateral recumbency. Multiple oxytocin boluses (10 to 20 U) were intravenously delivered to the sheep, and uterine contractions were monitored by TOCO readings. The body surface MRI markers were removed and replaced with unipolar active electrodes in corresponding locations. For the EMMI feasibility study, up to 256 body surface electrodes were connected to a BioSemi portable acquisition system (BioSemi) for noninvasive body surface electrical recording during oxytocin-induced uterine contraction about 15 to 40 min before surgery. The system sampled active electrodes at 2048 Hz with a 24-bit resolution. To accommodate the paralumbar surgical approach, body surface electrodes were retracted dorsally and ventrally. Lidocaine was infiltrated subcutaneously to create a local line block, and an incision was made in the right lateral paralumbar fossa. Upon entering the abdomen, the uterus was identified and exteriorized such that a proprietary electrode sock could be placed on the uterine surface. Oxytocin-induced uterine contractions were visually confirmed by two obstetricians and one veterinary surgeon. Unipolar pin-type active electrodes were inserted into electrode holders in the sock and in direct contact with the uterine surface. The sock was placed such that one column of electrodes was in line with the fetus’ spine. The positions of the electrodes were recorded as described in fig. S3 and further detailed in the Supplementary Materials. For sheep A and B, a cardiac pacing lead (controlled by the Medtronic 5375 Pulse Generator) was placed near one of the electrodes and in direct contact with the uterine surface. Next, the uterus was replaced in a normal anatomic position, and the abdomen was closed in layers. The sheep was repositioned into a prone position, and body surface electrodes were repositioned in their previous locations. Then, body surface and uterine surface electrodes were connected to a BioSemi portable acquisition system and simultaneously sampled for about 30 min. For assessing EMMI accuracy, the pacing signal was an electrical pulse train with about 0.5 or 1 s between pulses. For assessing EMMI robustness, all body surface electrodes were disconnected, and only uterine surface electrodes were used to record uterine surface potentials. After data collection was completed, the ewe and the fetus were euthanized, and all electrodes were removed, sterilized, and reused.

Bioelectric field computation

Under the assumption of an electro–quasi-static problem and the absence of an electrical source in the volume conductor, the Laplace equation (Eq. 1) was used to describe the electro–quasi-static field. The uterine and body surfaces were the boundaries encompassing the volume conductor as a multiple-connected domain. By defining the measured uterine surface electrical potentials as the Dirichlet boundary condition (Eq. 2) and using the known zero Neumann boundary condition (Eq. 3) on the body surface due to the electrical isolation effect of air, body surface potentials could be obtained by solving the Laplace equation (35)2φ(x)=0,xΩ(1)

Dirichlet conditionφ(x)=φU(x),xΓU(2)

Neumann conditionφ(x)n=0,xΓB(3)where Ω is the 3D volume between the body surface ΓB and the uterine surface ΓU. The volume Ω is assumed to be homogeneous. φ(x) represents the potential at location x. φU(x) represents the potential on the uterine surface. φ(x)n represents the normal derivative of the potential on the body surface, which equals zero because the body surface ΓB separates the conductive volume Ω and the nonconductive air outside of volume Ω. Similar to the validation work in ECGI (23, 34, 42), the boundary element method was used to discretize the Dirichlet and Neumann conditions and the Laplacian equation. After discretization, body surface potentials (ΦB) can be related with uterine surface potentials (ΦU) through a linear A, as Eq. 4. A is the transfer matrix encoding the body-uterus geometrical relationship. Because the boundary conditions were known on both boundaries, the computation was well posed, stable, and accurate (23).ΦB=AΦU(4)

Noise and geometric deformation

For EMMI robustness evaluation, we added Gaussian noise (10% in signal amplitude) into the computed body surface potentials (Eq. 5). We also added random Gaussian-distributed geometric error (three times SD = 1 cm) to both the uterus and body surface geometries.

EMMI inverse computation

In the inverse computation, under the same assumption as the bioelectric field computation, the Laplace equation governs the 3D electro–quasi-static field (Eq. 1). Body surface electrical activities were measured by a multichannel electrical mapping system using the Dirichlet boundary condition (Eq. 5) and the known zero Neumann boundary condition (Eq. 6) on the body surface.

Dirichlet conditionφ(x)=φB(x)+φN(x),xΓB(5)

Neumann conditionφ(x)n=0,xΓB(6)where φB(x) represents the potential on the body surface and φN(x) represents the Gaussian noise added to φB(x) on the body surface. To solve the inverse problem of EMMI, the MFS was used to discretize the Laplacian equation and Dirichlet and Neumann conditions as described by Eqs. 5 and 6. MFS was accurate for solving inverse bioelectric field problems in other systems (32). Because boundary conditions are known on the body surface and no boundary conditions are known on the uterine surface, the EMMI computation involves the inverse of a highly ill-conditioned matrix (32). To obtain a stable inverse potential solution on the uterine surface φ′(x), x ∈ ΓU, zeroth-order Tikhonov regularization was used (32, 64). The degree of regularization was determined by the CRESO method (23, 65). Specifically, we used a scaled mean CRESO parameter.

EMMI robustness evaluation

In this scheme (Fig. 3), up to 128 sock electrodes were used to directly record uterine surface potentials when the sheep was in labor. Oxytocin-induced uterine contractions were visually confirmed by two obstetricians (A.G.C. and J. Rhoades) and one veterinary surgeon (M.T.). We then combined the measured uterine surface potentials and the MRI-derived body-uterus geometry to forward compute the bioelectric field on the body surface. For this process, we used the boundary element method to discretize bioelectric field equations (34, 35, 42, 43, 66). We added (i) Gaussian noise (10% in amplitude) into the computed body surface potentials, (ii) Gaussian-distributed geometric error (up to 1 cm over all of the surface) to both the uterus and body surface geometries, or (iii) both 10% noise and 1-cm maximum deformation. Next, we used EMMI, which uses the MFS, to reconstruct uterine surface potential maps from the forward computed body surface maps (32). Last, we compared the measured and EMMI-reconstructed uterine data in terms of electrograms, potential maps, and activation isochrones.

Processing of potential maps, electrograms, and isochrone maps

Upon completing the EMMI inverse computation, uterine surface potential maps were generated, which display potential distribution over the entire uterine surface at a given time point. An electrogram provides temporal features of electrical activity at a local site on the uterine surface and was generated by assembling a time series of potential values at a given uterine site from the potential maps. An isochrone was generated by assembling local activation time (further descripted in the Supplementary Materials) of each uterine surface site during an observation window. The start of an observation window was selected as the time point when uterine electrical activity started to occur on a previously resting uterus, and the end of an observation window was selected as the time point when the uterus returned to electrical quiescence.

Statistical analysis

In this study, we used Pearson CC and RE to quantify accuracy of EMMI-reconstructed uterine surface potentials, electrograms, and isochrone maps. CC and RE have been well accepted and used in ECGI studies (25, 35, 67, 68).

The EMMI-reconstructed uterine surface potentials were compared to the measured uterine surface potentials by calculating the CC and RE. The equations defining CC and RE are as followsCC=i=1L(ViMVM¯)(ViRVR¯)i=1L(ViMVM¯)2i=1L(ViRVR¯)2(7)RE=i=1L(ViRViM)2i=1L(ViM)2(8)

CC is a statistical measure calculating the strength of the relationship between two variables. The range of CC values is between −1.0 and 1.0. A CC of −1.0 indicates a perfect negative correlation, whereas a CC of 1.0 indicates a perfect positive correlation. A CC of 0.0 indicates no relationship between the two variables. RE is a statistical measure of precision, defined as the ratio of the absolute error of a measurement to the measurement being examined. RE is dimensionless and expressed as a percentage.

For electrograms, we calculated the temporal CC and RE. L represents the number of sample points over time at which potentials were measured and reconstructed. ViM and ViR represent the measured and reconstructed potentials, respectively, at the ith sample point. VM¯ and VR¯ represent the temporal average of the measured and reconstructed potentials, respectively.

For potential maps and isochrone maps, we calculated spatial CC and RE. L represents the number of uterine sites at which potentials were measured and reconstructed. ViM and ViR represent the measured and reconstructed potentials, respectively, at the ith uterine site. VM¯ and VR¯ represent the spatial average of the measured and reconstructed potentials, respectively. In this analysis, reconstructed potential maps during contractions or pacing pulses were compared with corresponding measured uterine potential maps, and reconstructed isochrone maps were compared with corresponding measured isochrone maps. Wilcoxon rank sum test was performed to test the difference in the spatial resolution of measured data and EMMI reconstruction.

SUPPLEMENTARY MATERIALS

www.sciencetranslationalmedicine.org/cgi/content/full/11/483/eaau1428/DC1

Materials and Methods

Fig. S1. Measured and EMMI-reconstructed uterine surface electrograms in episodes 2 to 8.

Fig. S2. EMMI-reconstructed activation isochrone maps of nine oxytocin-induced contractions.

Fig. S3. Methods to generate sheep body-uterus geometry and to identify EMG burst clusters in electrograms.

Movie S1. Comparison of potential maps during pacing corresponding to Fig. 2A.

Movie S2. Comparison of potential maps during pacing corresponding to Fig. 2B.

Movie S3. Activation movie during observation A (corresponding to Fig. 6A).

Movie S4. Activation movie during observation B (corresponding to Fig. 6B).

Movie S5. Activation movie during entire episode 1 (including observations A and B).

Data file S1. Pacing potentials in Fig. 2.

Data file S2. Spacing of measuring electrodes and reconstruction points during pacing.

Data file S3. Reconstruction accuracy analysis of pacing pulses.

Data file S4. CCs and REs of episode 1 in Fig. 4A.

Data file S5. Reconstruction accuracy analysis of electrograms.

Data file S6. EMG potentials in Fig. 5.

Data file S7. Reconstruction accuracy analysis of potential maps.

Data file S8. Activation time sequences in Fig. 6.

Data file S9. Reconstruction accuracy analysis of isochrone maps.

Data file S10. Activation time sequences in Fig. 7.

Data file S11. Activation time sequences in fig. S2.

REFERENCES AND NOTES

Acknowledgments: We thank Y. Rudy and S. England for theoretical and technical advice. We thank Q. Wang for advice in MRI sequence setup. We thank J. Rhoades, D. Abendschein, C. Miller, A. Lewis, and J. Maurer for animal care and support. We thank D. Frank for editing the manuscript and J. Chubiz for coordinating the sheep delivery, MRI, and surgery. We thank Y. Dun, Z. Kisrieva-Ware, and Z. Wen for suggestions on analyzing and presenting data. Funding: This study was primarily supported by the March of Dimes [March of Dimes Prematurity Research Center; to principal investigator (PI) G.A.M.] and was also supported, in part, by grants from the NIH/National Institute of Child Health and Human Development (R01HD094381; to PIs Y.W. and A.G.C.), the NIH/National Institute of Aging (R01AG053548; to T. Benzinger and Y.W.), and the BrightFocus Foundation (A2017330S; to Y.W.). Author contributions: W.W., M.T., G.A.M., A.L.S., A.G.C., P.C., and Y.W. designed the EMMI study. Y.W., R.C.M., and P.K.W. designed the MRI protocol and optimized the MRI sequences. W.W., H.W., S.L., M.T., A.G.C., P.C., and Y.W. performed the sheep experiments. W.W., H.W., and Y.W. developed the EMMI software to analyze the data. P.Z. performed the statistical analysis. W.W., H.W., P.Z., M.T., A.G.C., P.C., and Y.W. wrote the manuscript. All authors revised the manuscript. Competing interests: Y.W., A.G.C., P.C., and A.L.S submitted U.S. provisional application no. 62/642,389 titled “System and Method for Noninvasive Electromyometrial Imaging (EMMI)” for the EMMI technology evaluated in this work. Y.W. serves as a scientific consultant for Medtronic and has NIH research funding. P.K.W. has research support from Siemens and research funding from Roche, Lilly, and NIH. R.C.M. has travel fund from Siemens. All other authors declare that they have no competing interests. Data and materials availability: All the data related to the manuscript are present in the main text and/or in the Supplementary Materials.
View Abstract

Stay Connected to Science Translational Medicine


Editor's Blog

Navigate This Article