Separation of Vibrational Cardiography signals by respiratory volume and phase using 1-dimensional Convolutional Neural Networks

Cardiovascular disease has been the leading cause of human mortality globally for years.Increasing cardiovascular health could reduce the frequency of cardiovascular disease and save lives, and preventative care has the potential to increase cardiovascular health.However, preventative care for cardiovascular disease is not ideal, most importantly lacking an established non-invasive, continuous method of cardiac monitoring.Non-invasive health monitoring could improve the application and efficiency of medical treatment and significantly reduce the frequency of cardiovascular disease-related mortality rates.Vibrational cardiography (VCG) has the potential to deliver non-invasive cardio-respiratory monitoring.VCG is the term given to a coupled seismocardiography (SCG) and gyrocardiography (GCG) measurement.VCG (along with its components, SCG and GCG) have been well studied and developed for cardiac monitoring.Moreover, the inherent effects of respiration on the VCG signal due to the proximity of the lungs to the heart have been studied as well.However, there is no established method of mitigating the respiratory variation in a VCG signal, thus reducing its efficacy as a cardiac monitoring tool.Approaches have been taken to filter out respiratory information from the VCG signal entirely, but studies have shown that this respiratory information could be useful for monitoring cardiovascular health.Instead, other approaches have been taken to separate VCG signals based on the respiratory phase or volume of the subject at the time they were recorded.This reduces respiratory variation in the signal without losing the potentially useful respiratory information altogether.The objective of this thesis is to take this separation approach, classifying VCG signals based on their respiratory volume and phase, specifically using 1-dimensional (1D) convolutional neural networks (CNN).1D CNNs are artificial neural networks which apply convolving filters to local features in one dimension.These networks are especially useful for analysing data in the temporal dimension and have been shown to have excellent performance in many signal processing domains, hence why they were chosen for this analysis.Data were collected from 50 subjects at McGill University, using an inertial measurement unit taped to the chest to obtain a VCG signal, and a spirometer to obtain a reference respiratory flow signal.Three classification objectives were examined: static respiratory volume, dynamic respiratory volume, and dynamic respiratory phase.For each objective, the cardiac cycles obtained from the VCG signals were manually split into one of two classes based on the respiratory flow signal and a 1D CNN was employed to classify these cardiac cycles based solely on their VCG information.I would especially like to express my gratitude to my two biggest and most influential collaborators, Yannick D'Mello and James Skoric, whose guidance from the time I was an undergraduate student led me towards this research and helped me at every point along the way with it.They were my introduction to the world of research and two of the biggest reasons I fell in love with this project.They taught me how to think like a researcher and how to navigate the often-difficult path of a master's student.Without their expert advice, assistance and sometimes criticism, this project would not be what it is today.I cannot express enough gratitude to these brilliant researchers for everything they have done for me.They began their journey with me as colleagues, and they have grown to become friends whom I will cherish forever.I would like to thank all of the other members of the Non-invasive Physical Activity Monitoring System (NiPAMS) team for their help in achieving my goals throughout this project.Namely, Ezz Aboulezz for his assistance with the acquisition system design and setup, Siddiqui Hakim and Angus McLean for their assistance with the data acquisition for the project, and Michel Lortie of the MDA corporation for supporting and trusting in this project throughout.I would like to thank my friends, both back home and in Canada, for being my support system throughout the course of this research.There are too many to name them all, but I would like to give special thanks to Sebastian Hunte, Brian Wood, Jaec Emtage-Cave, Isidora Conic and Stuart St. Hill for taking the time to review

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.