Applications of Machine Learning Algorithms in Fetal ECG Enhancement for E- Healthcare

Abstract

The fetal ECG (FECG) records the electrophysiological activity of the fetal heart and is considered as a useful tool for detecting fetal abnormalities during fetus growth. However, the recorded signals are often contaminated by various artifacts during acquisition. The vernix caseosa causes electrical isolation, which actually shrinks the signal amplitude, resulting in a poor signal-to-noise ratio of the FECG signal. Even the 3D morphology of FECG is quite challenging as the recorded FECG depends upon the position and shape of the uterus, fetal size, and presentation. The domain of FECG processing is attractive to many researchers as there is a lack of trustworthy exemplary databases with proficient interpretations and standard assessment tools for assessing the algorithms. Additionally, due to the lack of a 200sufficient number of cardiologist/radiologist and costly FECG recording device, this domain is gaining popularity in e-healthcare services. Machine learning (ML) algorithm, because of its adaptive learning approaches, find vast applications in FECG analysis. Algorithms like support vector machine (SVM), K-nearest neighbor (KNN), and Bayesian network are usually utilized for the detection of FECG from multichannel abdominal leads, whereas random forest algorithm helps in FECG acquisition. ML algorithms are also utilized for the classification of FECG signals using decision trees, linear and nonlinear regression models, principal component analysis (PCA) with Gaussian mixture models, and linear discriminant analysis. ML algorithms have been also used as a diagnosis tool to predict preterm births, fetal hypoxia, fetal heart rate, uterine contractions, and other chromosomal aneuploidies. Deep learning algorithms which are subsets of ML are gaining more popularity in this domain and are used for FECG signal enhancement, classification, and diagnosis purpose. In this chapter, firstly, we introduce FECG, their extraction methods, available databases for e-healthcare, and assessment tools. Secondly, different types and existing ML algorithms for FECG signals are explored. Finally, the outcomes of ML algorithms on fetal ECG enhancement for e-healthcare are presented.