UK researchers have developed a prototype wearable system for pregnant women, to detect foetal movements over long periods of time. The Nowlan and Vaidyanathan groups at Imperial College London, piloted the system in 44 pregnant women and managed to discriminate short bursts of movement (startles) from other general movements and breathing, as well as distinguish maternal artefactual movements. The team also set up an analysis framework and clinical protocol that can be used in future studies of upgraded monitoring systems (PLoS ONE doi: 10.1371/journal.pone.0195728). This is a first step towards the production of cheap, safe and wearable tech for monitoring foetal movements.

 Measures of foetal movement are often self-reported by the mother, typically by counting the number of kicks over a period of time. However, this is not a robust way to measure foetal activity; any reduction of movement may not be noticed, and if it is, it may be too late to intervene. A standardized measure would be invaluable for tracking foetal movements. Here, researchers provide a prototype device, worn by the pregnant mother, which can be used to collect data even during maternal movement. The system is low-cost and non-transmitting (and hence safe for the foetus).

Hardware development

The team collected 15 hours of data from 44 pregnant women (mean gestation of 31 weeks), using a custom-made inertial measurement unit consisting of eight acoustic sensors (used to detect vibrations from the foetus moving) and an accelerometer (which measured maternal acceleration in three axes).

Acoustic sensors, in a sealed chamber covered by a diaphragm, detect pressure changes when the outer membrane is perturbed by low-frequency vibrations caused by foetal movements. The research team also developed an ultrasound-compatible version, with sufficient field-of-view for the medical exam.

Detection and discrimination

The researchers correlated an ultrasound physician’s notes with three different movement types (breathe, general (whole-body) and startle). A detection was confirmed when observed movements during a 5 s window were in agreement with the sensors’ movement detection.

They found that 78% of startle (short, quick and directed) movements were detected correctly by the system. The detection was highly dependent on the proximity to a sensor, since startles may only be registered by a few sensors. The detection rates for whole-body movements and breathing were lower (53% and 41%, respectively), indicating that the system is only sensitive to startles.

The team also noticed a large variance between different scans; an inflated false positive rate meant that so far, the system cannot detect levels of foetal activity alone. The researchers suggest that the highly sensitive nature of the sensors makes it more susceptible to noise.

For the detected movements, the researchers employed principal component analysis (PCA) to reduce the dimensionality of the data. Machine learning classifiers were used to generate confusion matrices. Essentially, this procedure correlated the sensors’ detected movement to the physician’s observed movement, and hence output a percentage that described how well the classifier discriminated different movements.

Generally, startles were well discriminated from whole body movements (72.1% accurate) and breathing (66.6% accurate). However, the classifier did not discriminate well between whole-body movements and breathing.

Future prospects

The researchers plan to add more sensors to improve the specificity of foetal activity detection. Additionally, further optimization regarding placement and number of sensors should lead to improvements in detection and discrimination of movements. Given that detection of breathing and general movements were poor, the authors suggest that any wearable sensor would be unlikely to detect these movements.

This work has shown that foetal movements can be detected using acoustic sensors, and that the signal can be isolated from maternal artefacts captured by the accelerometer. For the first time, detection of startles and their discrimination from other movements using a wearable device has been described.