Identification of Discriminative Features for Uterine Contraction Detection in EHG Signals Recorded During Labor

The analysis of uterine electrical activity holds the potential to monitor pregnancy progression and uterine contractions with the aim to assess the risk of premature birth and to facilitate prompt medical action. This practice is performed using various monitoring tools, such as external palpation, tocography, intrauterine pressure measurement, etc. Among all of them, electrohysterography (EHG) is a promising technique due to its non-invasiveness. However, distinguishing uterine contractions from basal uterine activity within an EHG signal remains a significant challenge. This is especially due to the fact that the most adequate measures for detecting uterine contractions over an EHG recording are not yet fully understood and there is no consensus on the best approach to achieve this goal. To overcome the limitations mentioned above, this work aims to identify the features with the highest discriminative power for the automatic recognition of uterine contractions. Feature extraction was conducted on 6 EHG labor recordings, which were pre-processed and manually segmented by the help of a clinical expert into contraction and non-contraction intervals. A statistical analysis, followed by a visual inspection of its results, highlighted the superior performance of amplitude-related features (i.e. reflecting the energy of the signal). Indeed, Root Mean square, Teager Energy and Unnormalized First Statistical Moment have been identified as the features most effective for accurately identifying uterine contractions. These results hold promise for improving the development of automatic contraction detection methods, which can promote advanced pregnancy and fetal monitoring, ensuring timely and targeted interventions in case of adverse events.