Fetal ECG detection using time-frequency analysis with effective noise eliminations

During recording, Electrocardiogram (ECG) signals are often contaminated by various artifact noises, including white Gaussian noise, muscle activity, baseline wander, and power line interference. Therefore, pre-processing of ECG signals is essential to eliminate these artifacts and obtain efficient ECG features. Various attempts have been reported to eliminate ECG noise, including ECG signal denoising using wavelet transform (WT). However, many studies use direct ECG signals for the detection and classification of various abnormalities. The objective is to remove the different noises involved in fetal ECG using different filtering methods and wavelet decomposition. In the present work, we used various denoising techniques such as BlockJS, thresholding, BayesShrink, MinMax, and Sure. These methods have proven their efficacy in eliminating different types of noise. Subsequently, the discrete WT (DWT), a popular time-frequency analysis tool, was used to decompose these ECG signals into various subbands. Widely used features such as mean, skewness, kurtosis, variance, Hjorth parameters, and mean energy were extracted. Finally, a classification of abnormal and normal fetal ECG was performed. The effectiveness of the proposed method is assessed using a publicly available fetal ECG dataset from the CTU-CHB Intrapartum Cardiotocography Database, incorporating different types of noise across varying signal-to-noise ratios (SNRs). The experimental findings demonstrate that the present model outperforms existing methods and filtering, achieving a higher SNR and a lower RMSE. The experimental results showed that the 10dB WGN in BlockJS methods provided a maximum SNR of 21.25 with a classification rate of 95.90% from SVM. The proposed method performed well compared to existing methods, while preserving essential signal characteristics, making them highly suitable for ECG signal processing.