Project

# Bloomlife: improving prenatal health through longitudinal physiological monitoring at large scale

Goal: Bloomlife is a digital health startup focusing on helping expecting mothers have a healthy pregnancy. We are taking a data driven approach, combining data collected in clinical settings as well as consumer generated data to help shedding light on many poorly understood links between physiological changes naturally occurring during pregnancy, behavior and pregnancy outcomes.

Some of the applications we are working on are contraction detection (see our first product: https://bloomlife.com/), fetal movement monitoring, labour detection & maternal health.

We are particularly interested in the opportunities arising from crowdsourcing clinical research, by providing consumers with clinical grade tools and data and analyzing such data at a scale beyond what is possible in regular clinical studies.

All our published research is available in this project.

Methods: Machine Learning, Signal Processing, Electrocardiography, Preterm Birth, Preeclampsia, Fetal Movement, wearable technology, electrohysterography, labour detection, contractions

Date: 30 December 2013

0 new
11
Recommendations
0 new
7
Followers
0 new
33
11 new
1075

## Project log

This letter presents a 5-channel unipolar fetal electrocardiogram readout IC for monitoring the health of a fetus during pregnancy. Each readout channel includes an instrumentation amplifier, a programable gain amplifier and a successive approximation register ADC. A unipolar, common half branch reuse topology is used to achieve low noise, low power, low crosstalk between the channels high input impedance and high CMRR at the same time. Each channel achieves an input referred noise of 0.47 $\mu$ Vrms in 0.5 to 150 Hz, while consuming a power of 43.2 $\mu \text{W}$ . The 5-channel system provides a CMRR of 98 dB and an interchannel crosstalk rejection of 95 dB. The chip is implemented in a standard 55-nm CMOS process and occupies an area of 4.0 mm <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> . The whole chip, including five readout channels, leadoff detection, reference generation, autonomous data acquisition with on-chip sample storage and an interrupt-based serial peripheral host interface consumes a total power of 258 $\mu \text{W}$ .
Preterm birth (PTB), or birth before the completion of 37 weeks of pregnancy, is the leading cause of neonatal morbidity and mortality and the second-leading cause of death in children under the age of five. The importance of early intervention to prevent and reduce the impact of PTB is clear. However, no solution is currently available to easily allow for early detection of labour. The Bloomlife WISH system is a small wearable pregnancy monitor, worn on a pregnant woman's abdomen, that measures uterine electromyogram (EMG), maternal heart rate and movement, and uses that information to compute a probability of a woman being in labour. Here we report the preliminary validation of the labour probability estimated using this wearable pregnancy monitor.
Title* Early labour detection in laboratory and free-living conditions using combined electrohysterography and heart rate data Objective * Detection and management of complications such as preterm birth, could be improved by early labour detection. In our previous work, we showed that specific patterns in physiological data such as electrohysterography (EHG) and heart rate (HR) could be used to build predictive statistical models able to detect labour. In this work we highlight how physiological data can be discriminative of early term labour recordings when acquired both in laboratory and free-living conditions. Study Design* Accelerometry, EHG and HR data were collected under supervised laboratory conditions on 84 pregnant women using a wearable sensor designed to be attached to the abdomen using an adhesive patch as well as in free-living on 120 pregnant women. We extracted time and frequency domain features from EHG and HR data, as stronger, sinusoidal pattern arise on both data streams in correspondence of uterine contractions during labour. Features were used as input to a statistical model previously developed using recordings collected during labour at term and pregnancy as training set, to recognize labour and non-labour recordings. The statistical model was applied to preterm and early term recordings to assess the model's discriminative ability under those circumstances. Results * We report labour estimation probability for different conditions. Results showed that the probability of being in labour for recordings collected during the last 24 hours of pregnancy, when considering preterm or early term recordings, was consistently higher than the probability estimated for recordings collected outside of the last 24 hours. In particular, for recordings collected in supervised laboratory conditions, the mean probability of being in labour was 98% for actual preterm and early term recordings (combined dataset), while it was 0.1% for pregnancy recordings collected up to 24 hours before delivery. For free-living data, the mean probability of being in labour was 56% for actual early term recordings, while it was 27% for pregnancy recordings collected up to 24 hours before delivery. Conclusion * Our labour detection models demonstrated the ability to discriminate between early term and preterm labour recordings and non-labour recordings, using a combination of EHG and HR data. Our findings seem to indicate that the physiology of labour is similar for preterm and early term recordings, with respect to term recordings, as EHG and HR patterns were found discriminative of the conditions. Figure 1: Output probability of the labour detection models with respect to reference data, for early term and preterm recordings. Higher probability is consistently assigned to recordings collected during the last 24 hours before delivery.
Slides of my presentation at the 40th International Engineering in Medicine and Biology Conference (EMBC 2018), covering labour detection using physiological data (uterine and cardiac activity).

In this paper we show early evidence of the feasibility of detecting labour during pregnancy, non-invasively and in free-living. In particular, we present machine learning models aiming at dealing with the challenges of unsupervised, free-living data collection, such as identifying periods of high quality data and detecting physiological changes as labour approaches. During a first phase, physiological data including electrohysterography (EHG, the electrical activity of the uterus), heart rate (HR) and gestational age (GA) were collected in laboratory conditions for model development. In particular, data were collected 1) during simulated activities of daily living, aiming at eliciting artifacts and developing diagnostic models for free-living data 2) during pregnancy, including labour, aiming at developing labour probability models from clean, supervised physiological recordings. Machine learning models using datasets 1) and 2) were deployed in free-living, longitudinally, in 142 pregnant women, between week 22 of pregnancy and delivery. A total of 1014 hours of data and an average of 7 hours per person were collected. Output of the developed models was analyzed to determine the feasibility of detecting labour non-invasively using physiological data, acquired with a single sensor placed on the abdomen. Results showed that the probability of being in labour for recordings collected during the last 24 hours of pregnancy was consistently higher than the probability during any other pregnancy week. Thus, non-invasive labour detection from physiological data seems promising.
Our latest work titled Towards non-invasive labor detection: a free-living evaluation was accepted for publication at EMBC 2018. Abstract below, full paper will follow after the conference.
In this paper we show early evidence of the feasibility of detecting labour during pregnancy, non-invasively and in free-living. In particular, we present machine learning models aiming at dealing with the challenges of unsupervised, free-living data collection, such as identifying periods of high quality data and detecting physiological changes as labour approaches. During a first phase, physiological data including electrohysterography (EHG, the electrical activity of the uterus), heart rate (HR) and gestational age (GA) were collected in laboratory conditions for model development. In particular, data were collected 1) during simulated activities of daily living, aiming at eliciting artifacts and developing diagnostic models for free-living data 2) during pregnancy, including labour, aiming at developing labour probability models from clean, supervised physiological recordings. Machine learning models using datasets 1) and 2) were deployed in free-living, longitudinally, in 142 pregnant women, between week 22 of pregnancy and delivery. A total of 1014 hours of data and an average of 7 hours per person were collected. Output of the developed models was analyzed to determine the feasibility of detecting labour non-invasively using physiological data, acquired with a single sensor placed on the abdomen. Results showed that the probability of being in labour for recordings collected during the last 24 hours of pregnancy was consistently higher than the probability during any other pregnancy week. Thus, non-invasive labour detection from physiological data seems promising.

Labor is the physiological process during which the fetus is expelled from the uterus and is normally clinically diagnosed. However, the process leading to labor, typically involving an increase in contractions frequency and intensity can be monitored non-invasively using electrohysterography (EHG) and maternal heart rate (HR). Despite recent technological improvements, diagnosing or detecting labor outside of the hospital environment remains challenging due to much inter-personal variability and lack of data collected in free-living settings. In this abstract we present preliminary results of the deployment of the Bloomlife pregnancy tracker, a unique sensor able to capture EHG, HR, and accelerometer data. A sample of 51 women wore the sensor between pregnancy week 26 and 40 (or a subset of such period), including 12 labor recordings. We show that labor probability models developed in the lab can effectively provide information on the odds of delivering and could potentially be used in real-life situations to improve decision making as delivery approaches.
Our latest work was presented at BHI last month, the paper is titled: Towards personalized and non-invasive labor detection using bloomlife pregnancy tracker

Slides for our talk at the International Conference on Biomedical and Health Informatics (BHI 2017).

In this paper we propose a method combining electrohysterography and heart rate features to distinguish recordings collected during labour and during pregnancy. The paper was accepted for publication at the International Conference on Biomedical and Health Informatics (BHI 2017).

Continuation of our work on non-invasive fetal movement monitoring, accepted for publication at the International Conference on Biomedical and Health Informatics (BHI 2017). In this work we propose new methods to improve detection accuracy using a single sensor device.

Abstract accepted for publication at the Society for Maternal-Fetal Medicine conference, 2017. In this work we investigate the possibility to classify labour and non-labour recordings of electrohysterography and heart rate data.