Analysis of uterine EMG signals in term and preterm conditions using generalised Hurst exponent features. Issue 12 (1st June 2019)

• Record Type: Journal Article
• Authors: Punitha, N.; Ramakrishnan, S.
• Is Part Of: Electronics letters. Volume 55:Issue 12(2019)
• Page Start: 681
• Page End: 683
• Publication Date: 2019-06-01
• Subjects: obstetrics -- medical signal processing -- electromyography -- regression analysis -- fractals -- signal classification -- fluctuations -- Bayes methods -- sampling methods -- feature extraction; fluctuation analysis -- data‐balancing -- multifractal characteristics -- GHE features -- preterm uEMG signals -- LR classifier -- multifractal Hurst exponent features -- uterine EMG signals -- generalised Hurst exponent features -- uterine electromyography signals -- public database signals -- balanced feature contribution -- preterm conditions -- term conditions -- feature extraction -- synthetic minority over‐sampling technique -- classification algorithms -- Naive Bayes -- logistic regression -- signal classification; Electronics -- Periodicals; 621.381
• DOI: 10.1049/el.2019.0803
• Languages: English
• ISSNs: 0013-5194

Abstract

Abstract : An attempt has been made in this Letter to analyse term (week of gestation (WOG) >37) and preterm (WOG ≤ 37) conditions using uterine electromyography (uEMG) signals and generalised Hurst exponent (GHE) features. For this analysis, public database signals recorded from the surface of abdomen are considered. Multifractal detrended fluctuation analysis is performed on the signals and the GHE is calculated. From the exponent, seven features are extracted and data‐balancing based on synthetic minority over‐sampling technique is used to retain a balanced feature contribution by the term and preterm records. Two classification algorithms namely, Naive Bayes and logistic regression (LR) are employed to classify the signals. Ten‐fold cross validation approach is executed and the performance is validated using accuracy, precision and recall. The results show the uEMG signals exhibit multifractal characteristics and five GHE features are significant in distinguishing the term and preterm uEMG signals. The LR classifier gives the highest accuracy of 97.8%. Therefore, it appears that the multifractal Hurst exponent features in combination with LR classifier can be used as biomarkers for predicting the preterm or term delivery during the early stage of gestation.