Joint time-frequency analysis and non-linear estimation for fetal ECG extraction

This paper proposes a novel algorithm based on joint time-frequency analysis for extracting the fetal ECG using single-channel abdominal ECG. Stockwell transform is the time-frequency tool used in combination with Shannon Energy Entropy for identifying the maternal peaks. The single maternal beat is constructed as a Maximum likelihood estimator. The abdominal ECG free from the maternal component is again processed using S-transform to identify the fetal peaks. Masking and thresholding is done in the time-frequency domain to eliminate the high frequency and background noises. Inverse Stockwell transform combined with proposed region selective amplitude scaling (RSAS) is performed to denoise the FECG. Identifying maternal and fetal peaks in the time-frequency domain enables higher performance even in overlapping beats and further avoids the need for an explicit pre-processing approach. The proposed algorithm's performance is validated using Daisy database, Physionet Challenge 2013 Set-a dataset (PCDB), and Abdominal direct fetal ECG database (ABDFECGDB). Further to validate the consistency of proposed algorithm's performance, results obtained using real-time recordings acquired using Powerlab data acquisition hardware and data from Non-Invasive Fetal ECG Arrhythmia Database are included. The obtained results show the proposed algorithm's superiority with 97.37% of accuracy, 98.61% of sensitivity, 98.72% of positive predictive value, and 98.67% of F1 measure using PCDB database and 98.55% of accuracy, 99.16% of sensitivity, 99.38% of positive predictive value, and 99.27% of F1 measure using ABDFECGDB. Further, the algorithm shows consistent results using real-time data and arrhythmia dataset.

Automated summary

This paper presents a novel algorithm for extracting fetal ECG from single-channel abdominal ECG using joint time-frequency analysis. The algorithm shows superior performance in identifying the maternal and fetal peaks, and has been validated using various datasets.