Fetal Movement Detection With a Wearable Acoustic Device

Fetal movement counting is a simple method for monitoring fetal health. The counting can be self-performed; however, pregnant women require full attention during the process. It becomes difficult to perform other routine activities while counting. Therefore, wearable counting devices that can detect fetal movement and distinguish it from nonfetal movement are needed. In this work, an acoustic device was proposed for detecting fetal movement. The device’s dish was large, i.e., 31 mm, and therefore, a single device was sufficient for detection. Power spectral density (PSD) was used for creating the spectrogram of the moving signal. Four classifiers were used to identify fetal and nonfetal movements. The experiment was conducted on 12 pregnant women at their homes or workplaces, where body movement was allowed. The device was attached to the abdomen and recorded using a smartphone connected via Bluetooth. The total recorded time was 50.5 h. The signal amplitude of the fetal movement was 8.5 dB higher than that of the nonfetal movement. The detectivity of the device was 67.2%–75.2%. Lower detectivity was observed in subjects with small and very large abdominal circumferences. Gestation showed a negative correlation with maternal perception but no correlation with the device’s detection rate. The convolutional neural network classifier exhibited the highest performance, with a ${F}1$ -score of 95.2% and an accuracy of 94.9%.