Preterm birth prediction from electrohysterogram using multivariate empirical mode decomposition

Electrohysterogram (EHG) is an electrophysiological signal describing uterine contractions that can be non-invasively measured on maternal abdominal surface. This signal contains vital physiological and pathological information for assessing delivery abnormalities, such as preterm birth. However, extracting information that effectively characterizes the association with abnormal delivery from the weak EHG signal is challenging. We present a preterm birth predicting method using multivariate empirical mode decomposition (MEMD) algorithm that adaptively decomposes multichannel EHG signals into different intrinsic mode functions (IMFs). MEMD maintains spectral consistency across channels and avoids mode-mixing problems across IMFs due to its powerful fine-grained signal structure decoupling capability. On this basis, a total of 180 features were extracted from the IMFs and the final eight features were chosen using a two-step feature selection algorithm. A support vector machine (SVM) classifier was employed for decision-making. Specifically, cost-sensitive algorithm was used to solve the data imbalance problem. The proposed method was evaluated using 300 EHG recordings in TPEHG database. The results show that our method outperforms other state-of-the-art methods in terms of sensitivity (85.16%), specificity (96.54%), $$ (91.04%), accuracy (94.36%), and AUC (97.31%). This study provides a powerful tool with wide applications for preterm birth risk diagnosis in clinical obstetric.