Tracking evoked responses to auditory and visual stimuli in fetuses exposed to maternal high-risk conditions

Abstract

Magnetoencephalography (MEG) has been successfully applied to record fetal auditory (AER) and visual evoked responses (VER). In this study, we report the AER and VER development trajectory by tracking the evoked response detectability and latency from recordings starting at 27 weeks of gestation in pregnancies classified as high-risk. Fetal MEG and ultrasound recordings were performed on 158 pregnant women, and the total number of fetal auditory and visual tests conducted was 321 and 237, respectively. The overall evoked response analysis showed 237 AER (73.8%) and 164 VER detections (69.2%). The mean AER latency was 290.7 (SD 125.5) ms and the mean VER latency was 293.7 (SD 114.5) ms. The rate of decrease (95% confidence limits) in average AER and VER first-peak latency between 100–350 ms was 1.97 (−1.86, +5.81) ms/week and 1.35 (−3.83, +6.53) ms/week, respectively. This trend in high-risk fetuses conforms to the general trajectory of decrease in latency with gestational age progression, even though this decrease was non-significant, as reported in the case of normal growing fetuses. Although there was a significant difference in detection rates between male and female fetuses, this was not reflected in either latency values or the sensory modality applied. Further, the main factors that had the most significant effect on response detectability included the presence of intervening layers of adipose tissue between the fetal head and stimulus source and an increase in the maternal body mass index.

1. Introduction

The study of the fetal brain poses a number of unique challenges that have historically been restricted to postnatal investigation and extrapolation. During fetal life, disturbance of the fetal environment can potentially produce life-long effects (Gluckman, Hanson, Cooper, Thornburg, 2008). Thus, use of technology that is safe and, ideally, non-invasive is essential. While electroencephalography (EEG) and magnetoencephalography (MEG) recordings (Baillet, 2017, Cohen, 1972) have demonstrated their efficacy and safety in adults and even newborns (Dereymaeker et al., 2017; Chen et al., 2019) the maternal abdomen remains a formidable obstacle to assess fetal brain signals. Fetal EEG, in principle, is feasible from the surface of the maternal abdomen, but several factors including high impedance of the maternal skin weaken the tiny signals produced by fetal brain activity, making detection highly unreliable. Fetal magnetoencephalography (fMEG), on the other hand, measures magnetic fields rather than electrocortical potentials. (Eswaran, Lowery, Wilson, Murphy, Preissl, 2005). Magnetic fields permeate through body tissue and bones with no significant distortion and are, therefore, detectable over the maternal abdomen. Thus, fMEG is currently the only technology capable of non-invasively investigating fetal brain development by passively recording magnetic signals evoked by the fetal brain in-utero (Eswaran, Preissl et al., 2002; Eswaran, Wilson et al., 2002; Eswaran, Lowery, Wilson, Murphy, Preissl, 2004; Dunn, Reissland, Reid, 2015). Fetal functional magnetic resonance imaging has proven successful in assessing fetal brain activity; however, due to the low temporal resolution of this method, neuronal processes cannot be detected directly (Chen et al., 2019).

Transmission of the auditory stimuli across the abdomen to the fetus has been extensively studied (Gerhardt, Abrams, 1996; Lecanuet et al., 1998; Gélat et al, 2019) Fetal auditory responses have been successfully recorded into repeated pure tone stimuli in several fMEG studies (Holst et al., 2005; Kiefer et al., 2008, Linder et al., 2015). Similarly, simple short-duration repeated flash stimuli has been applied to obtain visual evoked responses (Eswaran, Wilson et al., 2002; Eswaran, et al., 2004). Additionally, paradigms investigating cognitive development such as habituation and mismatch response have been applied successfully using both auditory and visual evoked stimuli (Draganova et al., 2005; Draganova, Eswaran, Murphy, Lowery, Preissl, 2007; Hartkopf et al., 2016; Matuz et al., 2012; Sheridan et al., 2008; Lindsley, 1987). Several studies have shown that adverse in utero conditions in the prenatal stage of life could be an important indicator for later cognitive development (Lowery, Eswaran, Murphy, Preissl, 2006; Matuz et al., 2012; Muenssinger et al., 2013; Moraru et al., 2011; Morin et al., 2015; Schleger et al., 2014; Schleussner, Schneider, 2004; Sheridan, Matuz, Draganova, Eswaran, Preissl 2010; Sheridan, Draganova et al., 2010). Thus, tracking the fetal brain development can provide an insight into both normal and high-risk pregnancies with an adverse in-utero environment.

We have developed an optimized combination of recording parameters using auditory and visual stimuli to understand fetal neurological development trajectory. In the study, we have expanded the application of this approach to high-risk fetuses who were presented with auditory and/or visual stimuli (Eswaran et al., 2005). The overall goal of the study was to examine the potential feasibility of using auditory and visual evoked responses to track the neurodevelopment of high-risk fetuses for both research and potential clinical application. In high-risk pregnancies, the fetus is exposed to an adverse intrauterine environment that can influence its growth and development. These conditions can have a short-term effect and long-term consequences that fall in the framework of fetal origins of adult diseases (Nesterenko, Aly, 2009).

The high-risk population included fetuses that have been exposed to maternal conditions that could potentially lead to placental insufficiency thus compromising the fetal environment. The high-risk condition predominately consisted of intrauterine growth restriction, maternal preeclampsia and chronic hypertension. By performing recordings starting at 27 weeks of gestation, we report both the feasibility and challenges of recording fetal AER and VER, by tracking the evoked response detectability and related latency parameters. Further, we explored the association between uncontrollable factors such as fetal position and movement and maternal factors that can affect consistent levels of transmission of stimuli through the abdomen during recordings and response detection rates.

2. Material and Methods

2.1. Device

Data was collected using the SARA (SQUID Array for Reproductive Assessment) system built for the investigation of maternal-fetal physiology (Eswaran et al., 2005). The SARA system has 151 sensor channels arranged to fit the maternal abdomen and has a high spatial and temporal resolution. In Figure 1, on the top-left (a), a pregnant woman is shown leaning towards the sensor area on the SARA system inside a magnetically shielded room, where sound stimulus (Holst et al., 2005) is delivered to her abdomen. And on the bottom-left (b), light stimulus (Wilson, Adams, Murphy, Eswaran, Preissl, 2009) is delivered to the woman’s abdomen for visual evoked response studies.

Figure 1:

Pregnant women positioned in the SARA system. (a) Auditory stimulus delivered through a tubing. (b) Visual stimulus delivered through an optical fiber pad over the maternal abdomen. (c) Portable ultrasound prior to fetal MEG recording to determine the outer abdomen and head distance and documenting the intervening layers.

2.2. Participants

We recruited 158 high-risk pregnant women from the University Obstetrical Outpatient Clinic. Their “high-risk” status was determined by an obstetrician based on medical conditions and factors that potentially compromise fetal environment and health. The general demographics including body mass index (BMI), were recorded during their first visit.

The study was approved by the local Institutional Review Board (IRB), and informed consent was obtained from each participant. All the participants were tested for AER and/or VER at least once, and some of the women returned for additional tests (see Table 1). Out of the 158 women, 146 participated in AER testing and 112 in VER testing (100 participated in both, 46 AER only and 12 VER only). Return recordings were performed sequentially across GAs with no more than one fetal recording with a given modality performed on a single day. When both AER and VER were recorded on the same day then they were performed in a back to back sequence.

Table 1.

Participants, recordings, AER and VER detection.

Overall Gestational age (weeks)
	AER		VER
Mean (SD)	33.17 (2.59)		33.27 (2.66)
Median (Range)	33.29 (27.00--38.86)		33.29 (27.00--39.14)
Number of participants - visits/stimuli type/fetal sex/gestational age
	Total	Female	Male	Median GA (Range)
one visit	72	32	40	33.00 (27.00–36.14)
> one visit	86	40	46	33.37 (28.43–38.48)
Total Participants	158	72	86
	Total	Female	Male	Median GA (Range)
Both AER and VER	100	48	52	33.14 (27.57–38.48)
Only AER or VER	58	24	34	33.00 (27.00–37.50)
Total Participants	158	72	86
Number of tests and detection rate
		Total	Avg. per subject (SD)
# AER tests		321	2.2 (1.52)
# AER detections		237	1.62 (1.39)
AER response rate		73.83%
# VER tests		237	2.12 (1.62)
# VER detections		164	1.46 (1.35)
VER response Rate		69.20%
Sex differences in response rate
	AER		VER
Sex	Undetected (84)	Detected (237)	Undetected (73)	Detected (164)
Female	49 (58.3%)	109 (46.0%)	48 (65.8%)	78 (47.6%)
Male	35 (41.7%)	128 (54.0%)	25 (34.2%)	86 (52.4%)
Chi-square P value	0.052		0.0096

With a cross-sectional approach for AER testing out of 146 participants, 68 fetuses were tested once and 78 fetuses were tested more than once with 2–4 days delay between testing. Similarly, for VER testing out of 112 participants, 61 were tested once and 51 were tested more than once with 2–4 days delay between testing. The number of tests per fetus ranges from one to seven. This study included 75 female and 83 male fetuses. There were 51 African-American, 103 Caucasian, and three reported as Other. The Other category contained a mixed-race individual, a Hispanic, and a Pacific Islander.

2.3. Ultrasound Recordings

A detailed ultrasound exam was performed outside the shielded room in supine position for each of the recording sessions to determine fetal head location, movement (activity), and placental position. The head locations were classified over the maternal abdomen to be in vertical - upper (U) or lower (L) quadrant and horizontal - left (L) or right (R) quadrant with fetal head location in the middle of the quadrants represented as (M). For example, LL will denote lower left while UM will be upper midline with respect to maternal abdomen. The ultrasound recordings lasted up to 30 minutes in order to quantify fetal activities. The following activities were scored for their presence during the ultrasound period – extremity, gross body, breathing, mouth, and eye movements. In addition, if no significant movement was observed during this period, the activity was recorded as quiet sleep. The placental location was noted and classified as anterior, posterior, and fundal.

In the next step, the women were seated in the SARA system and a short ultrasound using a portable machine (Figure 1c) was performed to measure the distance between the outer abdominal wall and the fetal head. We classified the presence of the intermediate layers observed between maternal abdomen and fetal head that contributed to distance measure. The list of the layers included: amniotic fluid, adipose tissue, fetal body part (i.e. hand), placenta, maternal muscle, umbilical cord and uterine wall. The rationale in classifying the layers was based on the hypothesis that type of intermediate layer can affect the transmission of auditory and visual stimulus.

2.4. Fetal MEG Recordings

The stimulus delivery system was placed over the fetal head on the maternal abdomen and the women were asked to lean forward towards the SARA system for fMEG recordings. The auditory and the visual stimuli delivery sources were located outside the magnetically shielded room (Vakuumschmelze, Hanau, Germany). The sound stimulus was delivered to the inside of the shielded room through a 3.7-m length of flexible tubing that ended with an air-filled bag that was placed on the maternal abdomen. For the fetal measurements, the sound intensity measured at the end of the tube (outside of the bag) in the air was 110 dB (Holst et al., 2005). In the case of visual stimulus, the light was delivered through a 3.4 m fiber-optic cable that ended with a 3 cm × 5 cm woven panel (Stocker Yale, Salem, USA) that was placed over the maternal abdomen. The other end of the cable was divided into four sub-bundles, each illuminated by high-power LED arrays. The light output measured at the panel was 35 mW at 630 nm, red (Wilson et al., 2009).

MEG data was recorded at a sample rate of 312.5 Hz with a bandwidth of 0 – 100 Hz with recording session lasting up to an average of six minutes for each stimulation. The auditory evoked response protocol consisted of 100 ms tone bursts at 500 Hz (80%) – frequent tone and 1000 Hz (20%) –odd tone, with approximately 3 s inter-stimulus intervals (randomized +/− 0.5 s) delivered through a plastic tube. The odd tone was interspersed to avoid habituation. Visual stimulus duration was 500 ms with inter-stimulus interval of 3 s +/− 0.5 s. At the end of the fMEG session, we repeated the distance measure using the same procedure described above, and noted any position changes from the initial ultrasound.

2.5. Data Analysis

For post-processing, the direct current (DC) offset was removed and a high-pass filter at 0.5 Hz was applied for noise removal. The largest interference on fMEG data was from the maternal and fetal heart signals. From the raw MEG data, a maternal QRS complex between two successive fetal heartbeats was chosen as a template and the Orthogonal Projection (OP) operator technique was applied (Vrba et al., 2004). An equivalent approach was applied to remove fetal cardiac signals.

In the next step, AER and VER datasets were processed to determine the evoked response latencies. Data was bandpass-filtered between 0.5 and 10 Hz and split into one-second trials, between 0.2 seconds before the stimulus and 0.8 seconds after the stimulus. Any signal with an amplitude greater than 2 pT was excluded to eliminate any residual or movement artifacts. The fMEG data was then averaged and the latencies of evoked responses were determined. For AER latencies, the averages performed only on the frequent tone stimulus of 500 Hz were considered. A peak in at least 3 channels close to the fetal head’s location with amplitude greater than 5 fT and latency below 650 ms was accepted as an evoked response to determine the detectability of response. For trend analysis with GA, any peak latency between 100 ms and 350 ms was considered as the first-peak latency. This cut-off was determined from normative fetal MEG (Eswaran et al., 2005; Eswaran et al., 2004; Holst et al., 2005; Schleussner et al., 2004) and preterm neonatal evoked potential studies (Pasman, Rotteveel, de Graaf, Maassen, Notermans, 1991; Suppiej et al., 2015; Pryds, 1992; Taylor, Menzies, MacMillan, Whyte, 1987) where it has been shown that first evoked peak is a general broad peak that occurs around this latency range. The development of evoked response was further studied by dividing the recordings into the following three developmentally relevant GA Periods: Early (GA <32), Middle (32.≤ GA <35), and Late (GA ≥35).

Each recording was considered independent when testing for detection of an AER or VER. First-peak latencies for both AER and VER were analyzed with linear mixed-models regressions and ANOVAs. These mixed models include a random intercept effect along with the fixed effects of GA (continuous), GA Period, Fetal Sex, and the GA-Period-by-Fetal-Sex interaction. The random Intercept term models the intra-subject correlation among latencies measured at different times on the same fetus. The nature of the stimulation was accommodated by analyzing AER latencies as one outcome in one set of models, and by analyzing VER latencies as a different outcome in a different set of models. All tests were conducted using a two-sided alpha=0.05 significance level.

For AER and VER, the sessions in which an evoked response was detected were compared to sessions in which a response was not detected for differences in maternal and fetal characteristics. These comparisons employed Student’s t-test for the continuous-variable characteristics (BMI and the distances between outer abdominal wall and fetal head both before and after the fMEG recording), and the Chi-square test for the multi-category characteristics of fetal-head position and placenta location. For binary characteristics, we calculated the Chi-square test if n>5 in both cells and Fisher’s exact test for the others.

3. Results

3.1. Effects of maternal and placental variables on the detectability of the evoked responses.

Table 1 provides overall characteristics of recordings and detectability. The total number of auditory and visual tests conducted were 321 and 237, respectively, which resulted in 237 AER (73.8%) and 164 VER detections (69.2%). The average number of auditory tests per fetus was 2.2 with a detection average of 1.6 AERs per fetus. The average number of visual tests per fetus was 2.1 with a detection average of 1.5 VERs per fetus. The gestational ages (GA) ranged from 27 weeks to 38 weeks in the AER group, with an average GA of 33 weeks. In the VER group, GA ranged from 27 weeks to 39 weeks, with an average GA of 33 weeks.

Tables 2a and 2b provides a summary of the parameters investigated that could potentially affect the auditory and visual evoked responses respectively. These were extracted from the ultrasound-scoring sheet as described in the methods section. The table compares the frequency at which each factor is present in the detected and undetected groups along with the statistical parameters (df and P values). In the majority of fetuses, head position was localized in the lower midline quadrant (LM), followed by lower left quadrant (LL). Based on fetal-head position, the rates of detection versus non-detection were not significant for AER and VER. With regards to fetal movement, the most common occurrence was extremity and gross body movement, followed by breathing and mouth movements. There was a statistically significant difference between undetected and detected AER groups for movement of extremities (89.3% versus 95.4%, p = 0.048). In terms of the location of the placenta in the uterus, it was located anteriorly in majority large number of sessions. In AER and VER groups, actual placental location in the uterus did not have a significant effect on rate of detection.

Table 2a:

Analysis of factors that can affect AER

	Undetected AER (84)	Detected AER (237)	Test Statistic (df)	P-value
Fetal head position
Lower Midline (LM)	60 (71.4%)	174 (73.4%)	χ2=2.034 (df=3)	0.57
Lower Left (LL)	13 (15.5%)	24 (10.1%)
Lower Right (LR)	5 (6.0%)	17 (7.2%)
All Other	6 (7.1%)	22 (9.3%)
Fetal movement (activity)
Extremity	75 (89.3%)	226 (95.4%)	χ2=3.915 (df=1)	0.048
Gross Body	73 (86.9%)	208 (87.8%)	χ2=0.042 (df=1)	0.84
Breathing	62 (73.8%)	161 (67.9%)	χ2=1.010 (df=1)	0.31
Mouth	54 (64.3%)	172 (72.6%)	χ2=2.045 (df=1)	0.15
Eye	10 (11.9%)	30 (12.7%)	χ2=0.032 (df=1)	0.86
Sleeping	8 (9.6%)	12 (5.1%)	χ2=2.112 (df=1)	0.15
Placenta location
Anterior	30 (35.7%)	85 (35.9%)	χ2=3.955 (df=2)B	0.14
Posterior	12 (14.3%)	53 (22.4%)
Fundal	8 (9.5%)	16 (6.8%)
UndeterminedB	34 (40.5%)	87 (36.7%)
Distance between maternal abdomen and fetal head in a seated position prior to fMEG recording (cm)C
Mean (SD); NA	4.90 (1.68); 84	4.61 (1.81); 236	t=1.319 (df=318)	0.19
Distance between maternal abdomen and fetal head in a seated position post fMEG recording (cm)C
Mean (SD); NA	4.36 (2.29); 84	4.26 (2.15); 234	t=0.352 (df=316)	0.73
Intervening layers observed between maternal abdomen and fetal head
Amniotic fluid	54 (64.3%)	132 (55.7%)	χ2=1.878 (df=1)	0.17
Adipose	49 (58.3%)	83 (35.0%)	χ2=13.921 (df=1)	0.0002
Fetal part	19 (22.6%)	34 (14.3%)	χ2=3.079 (df=1)	0.079
Placenta	4 (4.8%)	35 (14.8%)	χ2=5.818 (df=1)	0.016
Maternal muscle	1 (1.2%)	7 (3.0%)	-----	0.69
Umbilical cord	0 (0.0%)	7 (3.0%)	-----	0.20
Uterine Wall	0 (0.0%)	13 (5.5%)	-----	0.024
Position change between the start and end of the fMEG recording
Position Change	15 (17.9%)	60 (25.3%)	χ2=1.902 (df=1)B	0.17
No Position Change	58 (69.0%)	148 (62.4%)
UndeterminedB	11 (13.1%)	29 (12.2%)
BMI (in kg/m2)C
Mean (SD); NA	35.6 (7.9); 76	31.7 (8.4); 197	t=3.490 (df=271)	0.0006

Test statistics are from Chi-square (χ²) tests or t-tests as indicated in the table cells; “-----” means Fisher’s exact test was performed. “df” denotes the test statistic’s degrees of freedom. All P-values are 2-sided; bold means P<0.05

Fetal head position, Placenta location, and Position Change During fMEG consist of mutually exclusive categories. All other factors consist of non-mutually exclusive yes/no indicators.

^A:Number non-missing.

^B:Undetermined placenta locations and undetermined position changes during fMEG were excluded from chi-square comparisons.

^C:Consist of continuous data.

Table 2b:

Analysis of factors that can affect VER

	Undetected VER (73)	Detected VER (164)	Test Statistic (df)	P-value
Fetal head position
Lower Midline (LM)	47 (64.4%)	117 (71.3%)	χ2=2.667 (df=3)	0.45
Lower Left (LL)	13 (17.8%)	20 (12.2%)
Lower Right (LR)	4 (5.5%)	13 (7.9%)
All Other	9 (12.3%)	14 (8.5%)
Fetal movement (activity)
Extremity	65 (89.0%)	155 (94.5%)	χ2=2.271 (df=1)	0.13
Gross Body	66 (90.4%)	145 (88.4%)	χ2=0.206 (df=1)	0.65
Breathing	54 (74.0%)	116 (70.7%)	χ2=0.262 (df=1)	0.61
Mouth	50 (68.5%)	115 (70.1%)	χ2=0.063 (df=1)	0.80
Eye	9 (12.3%)	17 (10.4%)	χ2=0.199 (df=1)	0.66
Sleeping	3 (4.1%)	5 (3.0%)	χ2=0.174 (df=1)	0.68
Placenta location
Anterior	26 (35.6%)	62 (37.8%)	χ2=4.823 (df=2)B	0.090
Posterior	9 (12.3%)	34 (20.7%)
Fundal	10 (13.7%)	11 (6.7%)
OtherB	1 (1.4%)	0 (0.0%)
UndeterminedB	27 (37.0%)	57 (34.8%)
Distance between maternal abdomen and fetal head in a seated position prior to fMEG recording (cm)C
Mean (SD); NA	4.93 (1.79); 72	4.63 (1.85); 164	t=1.157 (df=234)	0.25
Distance between maternal abdomen and fetal head in a seated position post fMEG recording (cm)C
Mean (SD); NA	4.88 (2.09); 73	4.37 (2.09); 162	t=1.733 (df=233)	0.084
Intervening layers observed between maternal abdomen and fetal head
Amniotic fluid	53 (72.6%)	105 (64.0%)	χ2=1.673 (df=1)	0.20
Adipose	41 (56.2%)	68 (41.5%)	χ2=4.392 (df=1)	0.036
Fetal part	10 (13.7%)	37 (22.6%)	χ2=2.496 (df=1)	0.11
Placenta	11 (15.1%)	24 (14.6%)	χ2=0.008 (df=1)	0.93
Maternal muscle	1 (1.4%)	4 (2.4%)	-----	1.00
Umbilical cord	1 (1.4%)	4 (2.4%)	-----	1.00
Uterine Wall	2 (2.7%)	7 (4.3%)	-----	0.73
Position change between the start and end of the fMEG recording
Position Change	9 (12.3%)	35 (21.3%)	χ2=3.115 (df=1)B	0.078
No Position Change	59 (80.8%)	113 (68.9%)
UndeterminedB	5 (6.8%)	16 (9.8%)
BMI (in kg/m2)C
Mean (SD); NA	36.3 (9.1); 67	30.8 (7.6); 145	t=4.572 (df=210)	<.0001

Test statistics are from Chi-square (χ²) tests or t-tests as indicated in the table cells; “-----” means Fisher’s exact test was performed. “df” denotes the test statistic’s degrees of freedom. All P-values are 2-sided; bold means P<0.05

Fetal head position, Placenta location, and Position Change During fMEG consist of mutually exclusive categories. All other factors consist of non-mutually exclusive yes/no indicators.

^A:Number non-missing.

^B:Undetermined placenta locations and undetermined position changes during fMEG were excluded from chi-square comparisons.

^C:Consist of continuous data.

Tables 2a and 2b also tabulate the results from the portable ultrasound performed before and after AER/VER recording sessions while women were seated on the SARA system. Results showed that the average distance from the outer abdomen to the presenting part of the head was around 4–5 cm. The layers that contributed to distance measures between the external maternal abdomen and the fetal head are listed in Tables 2a and 2b. The layers observed most often were amniotic fluid followed by adipose tissue. There was a statistically significant negative association between intervening adipose tissue and detection rate. In the AER subjects, intervening adipose tissue was present in 58.3% of the undetected group versus 35.0% of the detected group (χ²=13.92, df=1; p=0.0002). In the VER subjects, adipose tissue was present in 56.2% undetected versus 41.5% of detected (χ²=4.39, df=1; p=0.036). Placenta layer or fetal body part were also observed to be covering the fetal head in a number of sessions. While the head covered by placental layer did not significantly impact detection in the VER group, there was a significant positive association in the AER group. Only 4.8% of undetected AERs versus 14.8% of detected AERs had intervening placenta present (χ²=5.82, df=1; p=0.016). Maternal muscle and umbilical cord did not have any significant impact. There was a statistically significant negative effect of intervening uterine wall on AER detection, with 0% uterine wall present among undetected versus 5.5% among detected (Fisher’s exact p=0.024). Further, the results for maternal BMI demonstrates a strong statistically significant negative association between maternal BMI and rates of auditory and visual stimuli response detection. Average BMI was 35.6 in the undetected AERs versus 31.7 in the detected AERs (t=3.49, df=271; p=0.0006), and the average BMI was 36.3 in the undetected VERs versus 30.8 in the detected VERs (t=4.57, df=210; p<0.0001). Finally, the majority of fetuses did not have a significant change in position between the start and end of the recording, and this overall, did not impact the detection rate.

Table 3 presents the results of multivariate analysis of obstetric factors from Tables 2a and 2b that can impede the detection of an evoked response. When the evoked response was an AER, the presence of adipose tissue between the SARA and the fetal head was associated with a 2.23x increase in the odds of not detecting the response (χ²=7.07, df=1; p=0.0079); whereas,the presence of a placenta between the SARA and the fetal head was associated with a >79% decrease in the odds of not detecting it (χ²=8.76, df=1; p=0.0031). We note that the >79% decrease in non-detection rate is equivalent to a 4.82x increase in the odds of detecting an AER with an intervening placenta. Interestingly, BMI showed a strong association with AER detection rate in the univariate setting (Table 2a), but not in the multivariate setting (Table 3). Further, Table 3 shows that BMI was the only factor that was multivariately significant for VER. A 5-kg/m² increase in BMI was associated with a 1.51x increase in the odds of not detecting a VER (χ²=14.73, df=1; p=0.0001; Table 3), which is consistent with its behavior in the univariate setting (Table 2b).

Table 3:

Multivariate analysis of factors that affect evoked-response detection.

Effect on Odds of Undetected Response	AER Adjusted† Odds Ratio (AOR)	95% Confidence Limits on AER AOR, Lower Upper		AER P-value	VER Adjusted† Odds Ratio (AOR)	95% Confidence Limits on VER AOR, Lower Upper		VER P-value
Amniotic Fluid, Present vs Not	1.119	0.624	2.010	0.71	1.228	0.635	2.373	0.54
Adipose Tissue, Present vs Not	2.234	1.236	4.042	0.0079	0.878	0.416	1.852	0.73
Fetal Part, Present vs Not	1.452	0.675	3.116	0.34	0.506	0.224	1.142	0.10
Placenta, Present vs Not	0.207	0.073	0.588	0.0031	1.081	0.478	2.445	0.85
BMI in kg/m2, 5-unit increase	1.199	0.965	1.490	0.10	1.511	1.224	1.865	0.0001
GA in weeks, 1-week increase	0.978	0.886	1.078	0.65	1.033	0.925	1.154	0.56

^†Each factor in the table is adjusted for the presence of the other factors by means of multivariate logistic regression. The multivariate model for each outcome includes a GEE (generalized estimating equation) to model the dependency among different recordings from the same fetal-maternal pair.

3.2. AER and VER developmental trajectory across gestational age

Overall, the mean AER latency was 290.7(SD 125.5) ms and the mean VER latency was 293.7 (SD 114.5) ms. Figure 2 shows the spread of latencies across GA based on the first latency peak that was observed between 100–350 ms with male and female represented by different colored circles. The figure demonstrates that latencies of evoked response to auditory stimuli decreased with increasing gestational age. Average AER latency decreased at a rate of 1.97 ms/week across gestational ages 28 to 38 weeks with lower and upper 95% confidence limits (CLs) of −1.86 ms/week and +5.81 ms/week, respectively (F=1.06, df=(1,64); p=0.31). Similarly, Figure 3 shows the spread of VER latency over the recorded GA period. Again, we observed decreasing latencies with increasing GA in response to visual stimuli. Average VER latencies decreased at a rate of 1.35 ms/week across gestational ages 27 to 39 weeks with respective lower and upper 95% CLs of −3.83 ms/week and +6.53 ms/week (F=0.28, df=(1,41); p=0.60). While the numbers demonstrated decreasing latencies with increasing GA, these trends were not statistically significant. Furthermore, no statistically significant results were obtained from regression-based testing of first-peak latencies for fetal-sex differences or sex-by-GA interactions.

Figure 2:

The variation of AER with increasing gestational age with male and female fetuses represented by different colored circles. Each dot represents one recording per fetus at the given gestational age.

Figure 3:

The variation of VER with increasing gestational age with male and female fetuses represented by different colored circles. Each dot represents one recording per fetus at the given gestational age.

Further analysis classified GA into three periods; the mean ± SE at each period are shown in Figure 4 for both AER and VER. For AER, the interaction between GA period and fetal sex was not statistically significant (F=1.38, df=(2,61); P=0.26) and was removed from the analysis model. In the remaining main-effects-only model, average AER latency was 6.9 ms (3.0%) higher in male fetuses compared to female fetuses (t=0.75, df=63; P=0.46). Mean AER latency took successive downward steps from early to middle to late GA periods. The step sizes were 5.6 ms (2.4%) from early to middle (t=0.44, df=63; P=0.66) and 16.4 ms (7.1%) from middle to late (t=1.38, df=63; P=0.17) for a cumulative total decrease of 21.9 ms (9.4%) from early to late GA periods (t=1.68, df=63; P=0.097).

Figure 4:

The variation of AER and VER with gestational age split into three groups: Early, Middle and Late. The blue curve shows the AER trajectory, while the orange curve shows the VER trajectory. Filled circles (with error bars) denote the group means (with 95% confidence intervals) of the three GA periods.

Similarly, in the case of VER, the interaction between GA period and fetal sex was not statistically significant (F=0.43, df=(2,38); P=0.65), and was removed from the analysis model. In the remaining main-effects-only model, average VER latency was 17.6 ms (7.3%) higher in female fetuses compared to male fetuses (t=1.56, df=40; P=0.13). Mean VER latency took a slight step upward followed by a larger step downward from early to middle to late GA periods. The upward step size was 6.9 ms (2.9%) from early to middle (t=0.43, df=40; P=0.67), while the downward step size was 21.4 ms (8.9%) from middle to late (t=1.34, df=40; P=0.19) for a net decrease of 14.5 ms (6.0%) from early to late GA periods (t=0.84, df=40; P=0.41).

4. Discussion

To our knowledge, this is the only study that includes a relatively large cohort of more than 100 participants to undergo AER and VER recordings in order to examine the utility of MEG to track fetal neurodevelopment, especially in pregnancies classified as high-risk. In addition to advancing our scientific knowledge on fetal development, we specifically chose to focus on high-risk pregnancies based on its potential clinical benefit. Due to limited availability of MEG technology, its clinical application to high-risk population in tertiary care centers would be most beneficial. In this study, we were able to examine the detection rates and latency related parameters while laying emphasis on the fact that lack of detection is related to several uncontrollable factors. This is key especially in interpreting data in high-risk fetuses, since all the fetuses included in the study had conditions that could result in placental insufficiency, thus are more prone to potential neurological deficits. We believe that our work provides a good foundation to guide both scientific and clinical utility of this technology while identifying its limitations.

Once the detection rates were evaluated, we applied a two-pronged approach where we first performed a general latency analysis to determine if the times were within range and trends as observed in previous stimulus-related fetal MEG studies. In mixed-models analyses of latencies, we found for both AER and VER that the intra-subject correlations were weakly negative and statistically insignificant, and thus, consistent with zero correlation among different recordings from the same fetus. This supported the decision to treat different-day fetal tests as independent for purposes of evoked response detection. Secondly, we utilized our detailed ultrasound protocol to examine in depth the maternal fetal parameters that can affect response detection.

Although we have shown that fMEG can track brain development in high-risk fetuses, the rate of detected responses was around 70%, which was again within the range (50–90%) in previously reported fMEG studies in low-risk fetuses with a relatively low number of participants in each study (Eswaran et al., 2004; Holst et al., 2005; Draganova et al., 2005; Sheridan et al., 2008).

In terms of latency, the observed values, both for AER and VER, were similar for other reported studies as summarized by Dunn et al., 2015. Overall, the AER and VER latencies reported on low-risk fetuses are around 250 ms based on similar protocol adopted in this current study. The general hypothesis is that the evoked response latency should decrease with an increase in myelination as the fetal brain develops across gestation. On the other hand, it has also been shown that high-risk fetuses, especially the growth-restricted fetuses, exhibited longer age-adjusted AER latencies (around 360 ms) compared to normative fetuses (Kiefer et al., 2008; Morin et al., 2015). The results of this study on high-risk fetuses conforms to the general trajectory that fetal evoked response latencies decrease with increase in GA, although the decrease is not significant, as observed in the normal growing fetus (Holst et al., 2005).

Based on the timeline for the development of prenatal activities as described by Herschkowitz, 1988 - as determined by behavioral responses - startle response to vibroacoustic stimulation occurs around 24 weeks of GA, while response to light occurs around 28 weeks of GA. Although there are some differences in timeline for these functional developments, it is unclear if latencies averaged over GA will capture this effect other than providing a general trend. In order to account for maturational changes over GA, we classified the fetuses in to early, middle, and late GA groups. Comparative analysis between the groups mostly yielded decreasing trends in latency and none of them rose to level of significance. These results need to be examined further due to the limitation in categorization of high-risk pregnancies. Although all the fetuses tested were classified under potential placental insufficiency condition, there could be variation on how each condition can effect neurodevelopment. For example, IUGR may have a more profound effect than others such as maternal diabetes. In order to better quantify these effects, each of these conditions are being examined separately with matched, low-risk controls in our on-going fMEG studies.

As noted in Table 1, there was a significant difference in detection rates between male and female fetuses especially in the case of VER. But these differences were not reflected in the respective latency values. While there have been recent reports (DiPietro, Voegtline, 2017; Wheelock et al., 2019; Zavala, Ecklund-Flores, Myers, Fifer, 2019) showing differences in fetal heart rate variability and functional connectivity in the brain between male and female fetuses, our results of the effect of GA period, fetal sex, and their interaction-based testing of first-peak latencies for AER and VER for fetal-sex differences were not significant. Again this needs to be examined further, and as mentioned above, since latency trajectory may be different for each of the high-risk conditions, by combining them in one group we may have obliterated the potential effects of GA and sex.

One of the strengths of this study was that we utilized a detailed ultrasound protocol to document maternal fetal characteristics that can potentially affect the evoked response detection. The studies by Pasman, et al., 1991, in the early 90’s emphasized the need for determining the detectability of evoked response in preterm infants in order to provide appropriate clinical-neurophysiological correlation. Similarly, in the case of fetal evoked responses, the non-detectability of a response needs to be examined in detail. Especially in the case of fetus, the rate of response can be dependent on several uncontrollable factors such as fetal position and movement, and maternal factors that can affect consistent levels of transmission of stimuli through the abdomen (Schleger et al., 2014). In this study, we have shown that intervening layers such as tissue and muscle, could also have a significant effect on detection rates. In the AER group, univariate analysis showed statistically significant associations of detection rate were found with fetal-extremity movement, maternal BMI, and the presence of adipose tissue, placenta, and uterine wall interposed between the fetus and stimulus source. In the VER group, maternal BMI and intervening adipose tissue had statistically significant impacts on rate of detection. Biologically, we expect that adipose tissue is more likely to be present when BMI is higher, which means they should show a strong positive correlation with each other. When two factors are positively correlated in this manner, including them both in a multivariate regression model frequently makes one of them lose statistical significance towards the outcome of interest. For this reason, we think that the lack of multivariate significance of BMI towards AER is explained by the inclusion of the adipose tissue factor in the multivariate AER model. We also think that the same reason explains why BMI was multivariately significant towards VER, and adipose tissue was not, when both factors were in the multivariate VER model.

The statistically significant association between increased BMI and decreased rate of detection in both AER and VER groups suggests that women with higher BMI have increased adipose tissue that may depress or impede stimulus transmission, resulting in decreased stimuli detection by the fetus. This hypothesis is substantiated by the significant association found between presence of adipose tissue between the fetus and stimuli source and rate of detection in AER and VER groups.

In summary, this study specifically focused on high-risk pregnancies with protocols that were put in place to provide detailed ultrasound measures to evaluate the feasibility of fetal MEG application to this population. The results of the study gave an overall insight into development of auditory and visual evoked responses in high-risk fetuses while providing detailed analysis of influence of both maternal and fetal factors that need to be taken into account while interpreting the results. While we recognize that this analysis addresses only a portion of the factors that contribute to the viability of successful fetal brain recordings, we believe that this assessment at the fundamental level will allow us to obtain more reliable data from an inherently difficult study population. Integration of these factors is essential both in research and in clinical application of fetal MEG in order to provide reliable outcome information.