Non-Stationary Blind Source Separation and Advanced Signal Processing Techniques for Effective Fetal ECG Extraction

Fetal electrocardiogram (FECG) extraction from maternal electrocardiogram (MECG) signals presents a major obstacle in biomedical signal processing. In this paper Fast Independent Subspace Analysis (FastISA), Fast Independent Component Analysis (FastICA) and Join Approximate Diagonalization of Eigen Matrices(JADE) is used as for Blind Source Separation (BSS) with stern pre-processing steps like band pass filtering, wavelet denoising, orthogonalization and Wiener deconvolution. To evaluate the quality of separation Amari Index has been implemented achieving 0.19 for synthetic data using fastISA and 0.33 for real-world using JADE algorithm providing reliable operation for different situations as FECG extraction is crucial for its insights on fetal hearts electrical activity with non-invasive means to monitor and diagnose early during pregnancy. A predictive model based on Random Forest and a Neural Network along with R Peak Detection based on Pan-Tompkins algorithm has been implemented to predict pregnancy states and physiological metric for analyzing Mothers Health condition. Integration of predictive model and Signal processing techniques like FastISA and Pan-Tompkins algorithm provides a wider application of this approach in biomedical engineering for improved healthcare and diagnostics.