Bio Inspired methods for the extraction of fetal electrocardiogram

Abstract

This dissertation presents a comprehensive exploration of advanced algorithms for the ex- traction of non-invasive fetal electrocardiogram (NI-fECG) signals from abdominal electro- cardiogram (aECG) data, tackling the critical challenge of accurate fetal monitoring. The first study introduces a novel extraction method that combines the Grey Wolf Optimisation (GWO) with Sequential Analysis (SA). This innovative approach optimizes the parameters necessary for creating a template that effectively distinguishes the fetal electrocardiogram from the dominant maternal electrocardiogram (mECG) and other overlapping noise signals. The proposed GWO-SA method was evaluated on two real-world datasets—Labour and Preg- nancy—yielding remarkable results. Specifically, the extraction system achieved an average accuracy of 94.60%, an F1 score of 96.82%, a sensitivity of 97.49%, and a positive predictive value (PPV) of 98.96% for the Labour dataset. For the Pregnancy database, these metrics improved slightly, with an average accuracy of 95.66%, an F1 score of 97.44%, a sensitivity of 98.07%, and a PPV of 97.44%. These results underscore the method’s effectiveness in accurately detecting fetal QRS complexes, outperforming several state-of-the-art approaches. The second study conducts a thorough comparative analysis of five prominent population- based algorithms: Artificial Bee Colony (ABC), GWO, Moth Flame Optimisation (MFO), Particle Swarm Optimisation (PSO), and Whale Optimisation Algorithm (WOA). Each algo- rithm was employed alongside Sequential Analysis to extract the fetal electrocardiogram from the aECG data. The findings reveal that GWO, MFO, PSO, and WOA exhibited similar per- formance levels, achieving comparable extraction accuracy. However, the ABC-based system demonstrated instability and underperformed relative to the others. This analysis highlights the potential for further investigation into hybrid approaches and modifications to the ABC algorithm to enhance its extraction accuracy and stability in future applications. In the third study, the research pivots to the preliminary exploration of Empirical Mode Decomposition (EMD) combined with thresholding techniques aimed at enhancing the extrac- tion of fetal electrocardiography signals. This study underscores the importance of minimizing noise for the precise detection of key cardiac events, particularly the QRS complex, which is critical for assessing fetal health. While the results are still in the early stages, they indicate promising potential for EMD thresholding as an effective tool for noise reduction. This ap- proach lays the groundwork for further refinement and validation, aiming to improve clinical applications in fetal monitoring. Collectively, these studies provide significant insights into the development and optimiza- tion of algorithms for fECG extraction, demonstrating the efficacy of various methods and laying the foundation for future advancements in non-invasive fetal monitoring technologies. Ultimately, this research aspires to contribute to enhanced maternal and fetal health outcomes through improved monitoring techniques.