Development of a Novel Dataset and Tools for Non-Invasive Fetal Electrocardiography Research

This PhD thesis presents the development of a novel open multi-modal dataset for advanced studies on fetal cardiological assessment, along with a set of signal processing tools for its exploitation. The Non-Invasive Fetal Electrocardiography (ECG) Analysis (NInFEA) dataset features multi-channel electrophysiological recordings characterized by high sampling frequency and digital resolution, maternal respiration signal, synchronized fetal trans-abdominal pulsed-wave Doppler (PWD) recordings and clinical annotations provided by expert clinicians at the time of the signal collection. To the best of our knowledge, there are no similar dataset available. The signal processing tools targeted both the PWD and the non-invasive fetal ECG, exploiting the recorded dataset. About the former, the study focuses on the processing aimed at the preparation of the signal for the automatic measurement of relevant morphological features, already adopted in the clinical practice for cardiac assessment. To this aim, a relevant step is the automatic identification of the complete and measurable cardiac cycles in the PWD videos: a rigorous methodology was deployed for the analysis of the different processing steps involved in the automatic delineation of the PWD envelope, then implementing different approaches for the supervised classification of the cardiac cycles, discriminating between complete and measurable vs. malformed or incomplete ones. Finally, preliminary measurement algorithms were also developed in order to extract clinically relevant parameters from the PWD. About the fetal ECG, this thesis concentrated on the systematic analysis of the adaptive filters performance for non-invasive fetal ECG extraction processing, identified as the reference tool throughout the thesis. Then, two studies are reported: one on the wavelet-based denoising of the extracted fetal ECG and another one on the fetal ECG quality assessment from the analysis of the raw abdominal recordings. Overall, the thesis represents an important milestone in the field, by promoting the open-data approach and introducing automated analysis tools that could be easily integrated in future medical devices.