# Multi-layer self-calibrated algorithm for transabdominal fetal pulse oximetry: simulation and in vivo validation

## TL;DR

A new algorithm improves non-invasive fetal oxygen monitoring by accounting for maternal and fetal tissue layers, validated through simulations and sheep experiments.

## Contribution

A multi-layer self-calibrated algorithm that distinguishes maternal and fetal signals for more accurate fetal SpO2 estimation.

## Key findings

- The algorithm achieved a mean absolute error below 5% in simulations with high correlation to ground truth.
- In sheep experiments, the algorithm showed agreement with reference measurements (MAE = 10.3%).
- Algorithm performance is sensitive to accurate optical properties and tissue thickness inputs.

## Abstract

Transabdominal fetal pulse oximetry offers a promising approach to non-invasively monitor fetal arterial oxygen saturation (SaO2), potentially enhancing clinical decision-making and reducing unnecessary interventions during delivery. However, accurate estimation of fetal SaO2 (denoted as SpO2 when measured non-invasively) is complicated by the multi-layer maternal-fetal tissue structure, distinct maternal and fetal physiological signals, and inherently low fetal oxygen saturation levels. A multi-layer self-calibrated algorithm was developed by combining the multi-layer modified Beer–Lambert law with an analytical photon partial pathlength model. This approach distinguishes maternal and fetal tissue contributions, enabling more accurate fetal SpO2 estimation. Validation was performed using Monte Carlo photon simulations of multi-layer tissue geometries, where synthetic optical signals representing fetal cardiac pulsations were generated under two fetal depths and randomly varied maternal and fetal oxygen saturations and optical properties. Further validation was performed using in vivo sheep data, where fetal SpO2 values derived from transabdominal continuous-wave near-infrared spectroscopy measurements were compared against reference fetal SaO2 from CO-oximetry. In simulations, the algorithm achieved a mean absolute error (MAE) below 5% and a Pearson correlation coefficient (R) of 0.98 between estimated fetal SpO2 and ground truth fetal SaO2 when using optimal input parameters. In the sheep experiment, agreement with reference measurements was maintained (MAE = 10.3%, R = 0.91). However, algorithm performance was highly sensitive to accurate optical properties and tissue layer thicknesses inputs, which may be challenging to obtain in clinical settings. These results demonstrate proof-of-concept feasibility for the multi-layer self-calibrated algorithm in both simulated and in vivo conditions. While further refinement, particularly in optical property estimation and fetal depths in human pregnancies, is necessary, this work provides a foundational framework for the future clinical translation of non-invasive fetal SpO2 monitoring.