Non-invasive in vivo optical monitoring of human placental oxygenation

1 Direct assessment of human placental blood oxygenation can provide valuable information about 2 placental function and, potentially, detect dysfunction. Currently however, no bedside tools exist for non- 3 invasive monitoring of placental oxygenation. Here we report a continuous, non-invasive in vivo method 4 to probe placental oxygen hemodynamics using deep penetrating Frequency Domain Diffuse Optical 5 Spectroscopy (FD-DOS) with concurrent ultrasound (US) imaging. This multi-modal instrument facilitates 6 assessment of placental oxygenation properties from image reconstruction algorithms that integrate 7 anatomical US information about layer morphology with information from optics about functional 8 hemodynamics. Tissue phantom experiments, simulations, and human subject studies validate the 9 approach and demonstrate sensitivity to placental tissue located  5 cm below the surface. In a pilot 10 study (n=24), human placental oxygen hemodynamics are measured non-invasively during maternal 11 hyperoxia. Initial results suggest placental response to maternal hyperoxia may serve as a tool to detect 12 placenta-related adverse pregnancy outcome and maternal vascular malperfusion of placenta, weeks 13 before delivery. 14 direct in vivo assessment of human placental oxygenation (n=24). Specifically, we measured placental oxy- [𝐻𝑏𝑂 2 ] and deoxy-hemoglobin [𝐻𝑏] concentrations, or equivalently, total hemoglobin [𝐻𝑏 𝑇 ] concentration and oxygen saturation ( 𝑆𝑡𝑂 2 ) . We performed reproducibility and stability tests to characterize the technology, collected average tissue properties from each patient, and demonstrated detection of dynamic changes in placental oxygenation by varying maternal position and by performing maternal hyperoxia experiments. Notably, our pilot study shows that placental oxygen hemodynamics during maternal hyperoxia is significantly associated with 6 placenta-related adverse pregnancy outcome (APO) and with placental maternal vascular malperfusion (MVM), a 7 primary histopathologic pattern characteristic of placental dysfunction strongly associated with APO and with risk 8 of long-term disease 33,34 . The results demonstrate potential for non-invasive detection of placental dysfunction 9 and for generating improved clinical understanding of placental pathophysiology in vivo


15
[ ], and total-hemoglobin [ ] concentrations, and oxygen saturation ( 2 ) in placenta located ~5 cm below 16 the skin surface. Fig. 1 (a) shows key features of the custom heterodyne FD-DOS instrument. It employs three 17 lasers with wavelengths of 785, 808, 830nm. The output of each laser is radio frequency (RF) amplitude-18 modulated at 1 = 100 . A critical new technical feature of the instrument is its exceptional laser modulation 19 depth. To achieve this improvement, we divided the source driver signal into four sub-signals, amplified each sub-20 signal in multiple stages with low-noise linear amplifiers, and then recombined and impedance-matched the 21 amplified sub-signals for input to the laser drivers. Each laser's RF driver power was individually optimized to 22 achieve >90% light modulation depth, thereby increasing modulated diffusive wave amplitude and decreasing 23 (unmodulated) background diffuse light. As a result, measurement SNR is significantly better than previous work 24 (e.g., by >20dB for SDS = 8 cm), enabling long SDS measurements with smaller input powers (~35 mW).

7
These optoelectronics are fiber-coupled into a custom optical probe head, within which a commercial US probe 8 is also mounted. Optical source fibers in this probe offer 17 SDSs for patient measurements ranging, in the 9 present case, between 1 and 9 cm ( Fig. 1 (b)). During patient measurements, 10 source fiber locations are 10 chosen to optimize coverage over the anatomic regions of interest for each patient, and we scan sequentially penetration is required for placenta. The US transducer at the probe's center generates images ( Fig. 1 (c)) which 10 we utilize to segment target tissue into distinct layers that constrain optical reconstruction algorithms.

11
Three-layer modeling and reconstruction of placental properties. A critical algorithmic advance compared to 12 prior placenta research is our use of tissue layer morphology from US imaging to constrain the photon diffusion 13 tomographic inverse problem. In practice, we model the abdomen as three-layers: adipose, rectus/uterus, and 14 placental tissue. We approximate each layer as homogeneous and laterally infinite, but with thickness and depth 15 determined by US ( Fig. 1 (c)).

16
The measurement geometry and modeling are shown in Fig. 1 (c). is the SDS on the tissue surface. Given 17 the measured tissue layer thickness, as well as optical and physiological properties for each layer, standard 18 methods 35,36 are employed to solve the diffusion equation and generate predictions for the detected light fluence 19 rate on the tissue surface. Fig. 2(a) outlines our three-step reconstruction procedure (details provided in 20 Methods). Briefly, each step of the three-step reconstruction finds "best" tissue properties by minimizing the 21 difference between measured data and the predictions of diffuse optical tissue models of increasing complexity.

22
The STEP 1 assumes the underlying tissue is semi-infinite and homogeneous and employs the simplest 23 analytic model for optical property reconstruction. The STEP 2 utilizes estimates from step one as initial guesses 24 in a two-layer analytic diffuse optical tissue model. The STEP 3 utilizes estimates from step two as initial guesses 25 in a three-layer diffuse optical tissue model. In all steps, layer thicknesses are fixed by US imaging, but other 26 tissue properties are permitted to vary to find best estimates of tissue physiological and optical properties.

4
In two-layer phantom experiments, a solid phantom with fixed optical properties was positioned in a liquid 5 phantom, and the optical probe was set on the liquid surface ( Fig. 2 (c)). An absorption-titration experiment 6 evaluated instrument sensitivity, holding the over-layer liquid phantom thickness (3 cm) constant, while 7 incrementally increasing the absorption coefficient of the top layer. A depth-changing experiment tested sensitivity 8 to superficial layer thickness; here, the liquid phantom had fixed optical properties, and the superficial layer

11
Validation of three-layer reconstruction with finite element simulations. We generated simulated data using 12 a finite element simulation tool (TOAST) 37 that facilitated creation of a three-layer model with segmented optical 13 properties based on the layer morphology extracted from a patient ultrasound image ( Fig. 1 (c)). Since patient 14 layer interfaces are curved, we generated test data from curved patient layer interfaces ( Fig. 1 (c) left). For the 15 inverse problem, however, we assumed each layer interface to be flat ( Fig. 1 (c)

25
Adipose, rectus/uterus, and placenta layers were characterized by US and FD-DOS.

10
The 24 subjects were categorized into two groups according to their clinical pregnancy outcomes: NPO or 11 APO. Both 2 and 2 increased substantially in response to maternal hyperoxia for the NPO group ( Fig. 3 12 (c)). However, the same parameters in patients with APO showed a more blunted response ( Fig. 3 (d)).

13
Per patient placental histopathology, the same 24 subjects were categorized into two groups: NPP or MVM. We 14 observed significant, large and positive, r 2 and r 2 in the NPP group ( Fig. 3 (e)), but these same 15 parameters showed a blunted response in the MVM group ( Fig. 3 (f) were not associated with APO ( Fig. 4(a), Table 1 (a)). On the other hand Δ 2 was significantly reduced 22 in cases with APO compared to NPO, and Δ 2 was also reduced (marginally significant) ( Fig. 4(b), Table 1 (a)).

23
Similarly, we determined whether placental hemoglobin properties were significantly associated with MVM.

1
Direct and non-invasive methods to assess placental function in vivo and at the bedside do not currently exist.

2
The depth of the placenta below the skin surface, and the variability in the properties of the overlying layers, 3 presents significant challenges for optical diagnostics. Here we developed and reported on the performance of a 4 novel instrument and methodology that dramatically expands the capabilities of DOS to enable real-time, 5 continuous, dynamic monitoring of organs buried far below the tissue surface, such as the human placenta.

6
We demonstrated the instrument has sufficient dynamic range and SNR to perform measurements at long 7 SDSs (up to 10 cm) so that deeper layers can be optically interrogated. Moreover, by coupling FD-DOS 8 instrumentation with US imaging, we directly map the morphology of overlying layers of the abdominal wall and 9 uterus. This mapping permits multi-layer modeling of tissue properties that effectively isolates placenta 10 optical/physiological properties. We validated the methodology using tissue phantoms and finite element 11 simulations, and we carried out a pilot study of 3 rd trimester pregnancies.  and the other was prepared for driving amplitude modulation for one laser ('signal 1 '). Simultaneously, the 2 10 wave from the signal generator was also amplified, filtered, and divided (4-way splitter, ZB4PD-52-20W-S+, Mini-

11
Circuits) into four 2 waves, three were prepared for frequency-mixing with the three detected signals ('signal 2 ') 12 and the other was prepared for frequency-mixing with the reference signal ('reference 2 ').

13
The three of 'signal 1 ' were further amplified and input into laser controllers (CLD1011LP, Thorlabs), which

18
The three amplitude modulated laser diodes were fiber-coupled to an optical switch (MEMS 91545C4, Dicon), 1 which was in turn connected to the 10 source fibers (400 µm core, 0.5 NA, FP400URT-Custom, Thorlabs) on the 2 probe head (see main text Fig. 1 (b)). The optical switch sequentially cycled each laser diode through each 3 source position and also a "dark count position" (i.e., a cycle of 3 × 11 = 33 sequential measurements; 21 4 seconds per cycle). Of note, for the dark count measurement, no fiber was connected to the 11 th position on the 5 switch (i.e., no light was delivered to the tissue).

6
Multiply scattered light emerging from the tissue at the detector position was collected by a high-transmission

19
Note also, prior to computing and , the and for each wavelength at every source position were subtracted 20 by the corresponding and obtained from the dark count position in the same cycle. In summary, we 21 collect diffuse light waves from 10 source-detector pairs with source-detector separations (SDSs) ranging from ~1 22 to ~9 cm in the patient probe; these data enable the depth-dependent optical determination of tissue properties.

23
Three-layer photon diffusion model and associated Green's function. The human abdomen is multi-layered.

20
For multi-spectral fitting, we also assumed a Mie scattering model (below) for the tissue scatterers 15 , wherein 21 the scattering coefficient in layer is a power law function with scattering amplitude and scattering power .

22
Here, 0 = 700 is a reference wavelength chosen based on the range of the three wavelengths.

23
The first and very important step in the iterative search is initialization, wherein a reasonable estimated value 24 ̅ (0) is set. It is important that the initial guess is chosen to be reasonably close to the true value; otherwise the 1 iterative search may not converge to a meaningful solution. Main text Fig. 2 (a) schematically outlines our three-2 step reconstruction procedure for initialization and determination of placental hemodynamic properties.

3
In essence, each step of the three-step reconstruction finds "best" tissue properties by minimizing the 4 difference between measured data and the predictions of diffuse optical tissue models of increasing complexity.

5
The STEP 1 assumes the underlying tissue is semi-infinite and homogeneous; it utilizes short SDSs to derive an 6 initial estimate for properties of the near-surface region and the full set of SDSs to derive an initial property 7 estimate for the whole region. For predictions, a standard semi-infinite homogeneous medium analytic solution is 8 used in step one.

9
The STEP 2 utilizes estimates from step one as initial guesses in an analytic two-layer diffuse optical tissue

20
Experimental validation/characterization with tissue-simulating phantoms. We characterized FD-DOS 21 instrument performance using tissue-simulating phantoms. The simplest tissue phantoms were comprised of 22 water, ink for absorption, and 20% Intralipid (Baxter) for scattering. Briefly, the detector fiber was fixed at the liquid 23 surface and a source fiber was physically translated in the same surface plane with SDSs ranging from 6.2 cm to 24 10 cm using a translation stage (see main text Fig. 2 (b)). To good approximation, the coupling coefficients in demonstrate good SNR at SDSs up to 10 cm (see main text Fig. 2 (b)) and accuracies in the range of 3% to 9%.

4
The over-layer liquid phantom thickness was held constant at 3 cm, the over-layer scattering coefficient was held 5 constant ( , ′ = 9.10 −1 for 1 = 785 ), and the absorption coefficient was incrementally increased in the top 6 layer from , = 0.08 to 0.13 −1 . Results are given in Extended Data Table 1 (left). A depth-changing 7 experiment tested sensitivity to superficial layer thickness (see main text Fig. 2 (c)). Here, the liquid had fixed 8 optical properties ( , = 0.13 −1 , , ′ = 9.10 −1 for 1 = 785 ), and the superficial layer thickness was 9 increased from 1.5 to 3.0 cm. Results are given in Extended Data

18
In total, n=24 subjects participated in this study. Detailed information about the subjects are provided in

19
Extended Data Table 3. Note, data from two other subjects were excluded because their signals were either too 20 small (tissue optical absorption coefficient was very large, > 0.2 −1 ) or too unstable (due to large fluctuations 21 during baseline period).

22
Measurement reproducibility was evaluated using the Intra-class Correlation Coefficient (ICC). We measured 23 the hemoglobin properties multiple times at the same placental location in 19 subjects.

24
The stability test (n=24) results were represented by the standard deviation (S.D.) during the continuous 10 25 frames measurements (Extended Data Fig. 2 (a) frames both before and after the maternal tilt. Extended Data Fig. 2 (b)

25
[ 2 ] were calculated using the final 4 frames of the baseline period. Δ 2 , Δ , and Δ 2 were defined as 26 the difference between these mean baseline values and the "peak" values of the 4-frame window during maternal hyperoxia wherein maximum 2 occurred. Each subject (N=24) was then categorized into two groups based on 1 pregnancy outcome: NPO or APO. We observed a significantly larger Δ 2 and Δ 2 in response to maternal 2 hyperoxia in the NPO group compared to a more blunted response in the APO group (see main text Fig. 4 (b)).

3
Similarly, when analyzing placental histopathology as the outcome of interest, we observed significant (large) 4 Δ 2 and Δ 2 in the NPP group compared to a blunted response in the MVM group (see main text Fig. 5 (b)).

5
Wilcoxon rank sum tests were performed to calculate the p-values for comparison of different variables between 6 NPO vs APO groups and NPP vs MVM groups.

7
We also analyzed the static and dynamic hemoglobin properties of the adipose and rectus/uterus layer from 8 our three-layer reconstruction (Extended Data Table 4). (Note, due to very thin adipose or rectus/uterus layer 9 thickness, 4 of the 24 subjects were processed with two-layer model reconstruction instead of three-layer model 10 reconstruction; therefore, we excluded these 4 subjects in the statistical analysis of adipose or rectus/uterus 11 layer.) The resulting data provide in vivo evidence demonstrating the variability of the optical properties of the 12 overlying layers, thereby underscoring the importance of the multi-layer modeling to separate layer responses.

13
FD-DOS/US is critical for quantitative capture of placenta response; without the multi-layer model and associated 14 instrumentation, estimates of placenta response would be contaminated by signals from the other layers.

15
Uterine Artery Doppler Pulsatility Index (UtA PI). Each uterine artery was identified using a transabdominal 16 C1-5 ultrasound probe (GE Healthcare) via power Doppler mapping. Pulsed wave Doppler was then used to 17 obtain three similar consecutive waveforms. PI was defined as the difference between peak systolic and end 18 diastolic velocities divided by the mean velocity. The mean PI of the two uterine arteries was used for analysis. No 19 association was found between UtA PI and APO or MVM (main text Table 1 (c)). Furthermore, when including 20 UtA PI as a covariate, the associations between Δ 2 and Δ 2 and our outcomes (i.e. APO and MVM) 21 remained apparent (Extended Data Table 5).

22
Statistical Analysis. Statistical analysis was performed using MATLAB 2019a. Depending on the data-type, 23 results are presented as mean (S.D.) and median (IQR). ICC in the reproducibility experiment was calculated by 24 dividing the random effect variance by the total variance. P-values for studying correlations between different 25 variables and placental dysfunction were obtained by two-sided Wilcoxon rank sum test, which is a nonparametric 26 test for two populations when samples are independent. P values for studying correlations between nulliparity and 27 28 placental dysfunction were obtained by two-sided Fisher's exact test. P-values for studying the before/after 1 difference in the maternal left tilt experiment were calculated by two-sided paired sample t test analysis. Binary 2 logistic regressions were also performed to study the correlation between APO/MVM and Δ 2 or 2 but with 3 control of other variables: UtA PI, placental depth ( ), and pre-gravid BMI. We carried out this analysis for 4 completeness with caveats that the sample size is small and that different pairs of variables might be partially 5 correlated (and if so, that future inclusion of interactions in the statistical models is desirable). Extended Data 6 Table 5 shows the resultant P-values of Δ 2 and Δ 2 for prediction of APO or MVM from the binary logistic 7 regression models. The results confirm that there remained a trend towards significant association between 8 optically-derived hemodynamic properties and our outcomes of interest. For the future, a larger sample size will 9 permit more sophisticated statistical analyses that explore the effects of possible confounding variables and that 10 generate composite metrics with improved specificity and sensitivity.

11
Data Availability

12
The data that support the findings of this study are available in the main text of the paper and in its supplementary 13 materials. In addition, all raw data are available from the corresponding author upon request.
14 Code Availability

15
The custom code employed for processing the optical data and for performing the statistical analysis are available 16 online at https://github.com/LynnWong34. The LabVIEW code and simulation code are also available from the 17 corresponding author upon request. Note: Parameters were summarized as median (IQR).

Extended Data
*Four subjects were analyzed with 2-layer rather than 3-layer model due to thin adipose or muscle layers.

3
Extended Data Note: Binary logistic regression was applied to estimate the association between APO (or MVM) and Δ or ΔHb 2 with control of other variables: UtA PI, placental depth ( ), and pre-gravid BMI.