FastICA Algorithm Based on Independent Component Analysis for Maternal and Fetal ECG Signal Separation

Main Text: In the field of biomedical signal processing, separating maternal and fetal electrocardiogram (ECG) signals presents a challenging task. Since fetal ECG signals are typically masked by stronger maternal ECG signals, traditional filtering methods often struggle to effectively extract clear fetal ECG components. The FastICA algorithm based on Independent Component Analysis (ICA) provides an effective technical solution to address this problem.

The core concept of the FastICA algorithm involves achieving blind source separation of signals by finding a set of statistically independent components. The algorithm assumes that mixed signals are linear combinations of multiple independent source signals, and estimates these independent components by maximizing non-Gaussianity. When processing maternal and fetal ECG signals, FastICA can decompose the mixed signals collected from abdominal electrodes into independent components representing maternal and fetal ECG signals. In code implementation, this typically involves creating a mixing matrix and applying whitening transformations using functions like `pca()` or `zca_whitening()`.

Compared to traditional ICA algorithms, FastICA employs a fixed-point iteration optimization strategy that significantly improves convergence speed. It uses an approximate Newton iteration method to find the directions of independent components, avoiding the slow convergence issues of traditional gradient descent methods. During implementation, the algorithm first performs centering and whitening preprocessing on the mixed signals, then estimates the non-Gaussianity of independent components through nonlinear functions (such as tanh), ultimately obtaining separated signal components. Key programming steps include implementing the `fastica()` function with whitening preprocessing and nonlinearity optimization using `np.tanh()` or similar activation functions.

In practical applications, the FastICA algorithm can effectively separate maternal and fetal ECG signals and demonstrates certain robustness to noise. However, its separation effectiveness is influenced by signal quality, electrode placement, and algorithm parameter settings. Therefore, it typically requires combining preprocessing and postprocessing steps to further enhance signal separation performance. Code implementation should include signal quality checks using `scipy.signal.filter()` and parameter optimization loops.

It should be noted that ICA-class algorithms inherently possess order uncertainty characteristics, requiring the identification of separated components through prior knowledge of ECG signals. Despite this, FastICA remains one of the more effective methods for maternal-fetal ECG separation currently available, providing important technical support for prenatal monitoring. Implementation typically includes component identification algorithms using ECG waveform templates and correlation analysis with `scipy.correlate()`.