Classification of Abdominal Fetal Electrocardiogram Recordings using Karhunen-Loève Decomposition
Maternal abdominal ECG acquisition is a non-invasive technique not usually common in clinical practice but with a growing interest since it could provide enough information to explore the cardiovascular condition of the fetus after the 20th week of gestation. Since the maternal and fetal cardiac activities are combined in the ECG signal, several challenges have to be surmounted to obtain a useful fetal ECG signal from the maternal abdominal ECG like the overlapping observed in time and in frequency of both components and the low signal-to-noise ratio of the signal. In the present study, the maternal and fetal activities were exploited through the decomposition of maternal abdominal ECG recordings using the Karhunen-Loève (KL) transform. Then, several set features extracted from the KL decomposition and from the ECG signal were combined in support vector machines classifiers to detect the presence and absence of a maternal and/or a fetal QRS complex from ECG segments of 250 ms duration. Results show that a fetal QRS complex was well detected when there was not overlap with a maternal QRS (sensitivity 88.8%, specificity 97.1%) while detection diminished somewhat when overlapping was present (sensitivity 90.7%, specificity 75.9%).
KeywordsKarhunen-Loève decomposition abdominal fetal electrocardiogram support vector machine eigenvector analysis biosignal processing
Unable to display preview. Download preview PDF.