C D OMPUTATIONAL ETECTION OF E A LECTROPHYSIOLOGICAL BNORMALITIES H C IN YPERTROPHIC ARDIOMYOPATHY FOR R S ISK TRATIFICATION Aurore Lyon St Hugh’s College Computational Cardiovascular Science Group Department of Computer Science University of Oxford Michaelmas 2017 This thesis is submitted to the Department of Computer Science, University of Oxford, for the degree of Doctor of Philosophy. This thesis is entirely my own work, and, except where otherwise indicated, describes my own research *** To my parents Avec une pensée pour Claude et Philippe *** i ii Aurore Lyon Doctor of Philosophy St Hugh’s College Michaelmas Term 2017 COMPUTATIONAL DETECTION OF ELECTROPHYSIOLOGICAL ABNORMALITIES IN HYPERTROPHIC CARDIOMYOPATHY FOR RISK STRATIFICATION ABSTRACT Hypertrophic cardiomyopathy (HCM) is a common cardiac genetic disease and a major cause of sudden cardiac death (SCD) in young adults. While most patients remain asymptomatic, others suffer from SCD triggered by ventricular arrhythmias. An accurate detection of these high-risk patients in order to provide them with appropriate treatment is therefore essential, and it remains a challenge as current electrocardiogram (ECG) biomarkers are not specific. In this thesis, we develop computational methods for the analysis and interpretation of electrophysiological clinical data in order to investigate the diversity of HCM phenotypes with the ultimate aim to improve risk stratification in HCM. First, by combining computational clustering and mathematical modelling, we identify four distinct ECG phenotypes exhibiting differences in hypertrophy distribution and risk of arrhythmia. The group with primary repolarization abnormalities and coexistence of apical and septal hypertrophy shows a higher HCM Risk-SCD score compared to other groups. Second, we explore the influence of structural and electrophysiological mechanisms on the ECG to explain the four HCM phenotypes identified, by using a whole-body personalized 3D computer simulation framework. We show that apico-basal repolarization heterogeneities explain the T wave inversions in the high risk group, and that an abnormal Purkinje system may explain the QRS abnormalities in another group. Finally, we further investigate the presence of repolarization biomarkers in HCM, specifically action potential alternans, and show that HCM cardiomyocytes do not overall exhibit more action potential alternans compared to controls, which may suggest other whole-organ mechanisms responsible for T wave alternans. Overall, this thesis contributes towards a better understanding of HCM phenotypic heterogeneity and the improvement of individual patient management. iii Publications The publications resulting from the work described in this thesis are listed below. A. Lyon, A. Mincholé, J P. Martínez, P. Laguna and B. Rodriguez. Computational techniques for ECG analysis in light of their contribution to medical advances. J. R. Soc. Interface. 2018 15 20170821; DOI: 10.1098/rsif.2017.0821. A. Lyon, R. Ariga, A. Mincholé, M. Mahmod, E. Ormondroyd, P. Laguna, N. de Freitas, S. Neubauer, H. Watkins and B. Rodriguez. Distinct ECG phenotypes identified in hypertrophic cardiomyopathy using machine learning associate with arrhythmic risk markers. Front Physiol. 2018; 9:213. DOI: 10.3389/fphys.2018.00213 A. Lyon, A. Mincholé. E. Passini and B. Rodriguez. Investigation of the Presence and Mechanisms of Action Potential Alternans in Hypertrophic Cardiomyopathy. Computing in Cardiology. 2017. A. Lyon et al. Investigation of the effect of structural abnormalities on the ECG in hypertrophic cardiomyopathy patients using MRI-based computer models of the human heart and torso. Submitted to Europace. A. Lyon, R. Ariga, A. Mincholé, P. Laguna, N. de Freitas, S. Neubauer, H. Watkins, B. Rodriguez. Risk stratification in hypertrophic cardiomyopathy based on QRS and T wave morphological biomarkers identifies three phenotypic subgroups. European Heart Journal, 37 (Abstract Supplement), 412-413. 2016 A. Ghetti, V. Grau, S. Harmer, I. Kopljar, P. Lambiase, H. Rong Lu, A. Lyon, A. Minchole, A. Muszkiewicz, J. Oster, M. Paci, E. Passini, S. Severi, P. Taggart, A. Tinker, J.-P. Valentin, A. Varro, M. Wallman, X. Zhou, Human-based approaches to pharmacology and cardiology: an interdisciplinary and intersectorial workshop. Europace 18(9). 2016. A. Lyon, A. Mincholé, R. Ariga, P. Laguna, N. de Freitas, S. Neubauer, H. Watkins, B. Rodriguez. Extraction of morphological QRS-based biomarkers in hypertrophic cardiomyopathy for risk stratification using L1 regularized logistic regression. Computing in Cardiology. DOI: 10.1109/CIC.2015.7408573. 2015 iv Conference presentations The work described in this thesis was presented at the following conferences: CMR2018, Barcelona, February 2018 (invited talk) Computing in Cardiology, Rennes, September 2017 (talk) BHF Centre for Research Excellence Symposium – Oxford, September 2017 (invited talk) European Medical and Biological Engineering Conference (EMBEC), Tampere, June 2017 (poster) BHF Centre for Research Excellence Symposium – Oxford, September 2016 (poster) European Society of Cardiology (ESC) Congress, Rome, August 2016 (talk) Young Investigator Award Competition at Computing in Cardiology, Nice, 2015 (talk) BHF Centre for Research Excellence Symposium – Oxford, September 2014 (poster) Prizes The work described in this thesis was awarded the following prizes: Best talk, Oxford and Cambridge Women in Computer Science Conference, Oxford, March 2017 Winner of the Young Investigator Award, Computing in Cardiology, Nice, September 2015 Honorary mention for Best Poster, BHF Centre for Research Excellence Symposium, Oxford, September 2014 v ACKNOWLEDGEMENTS First of all, I would like to thank my supervisors, Professor Blanca Rodríguez and Dr Ana Mincholé, for all their support over the last three years. Ana, thank you for being available, supportive, reassuring and optimistic every single day. Most of this work would never have been done without your help! Blanca, you have been an incredible model to look up to and I have learnt so much by your side. Thank you for supporting me through failure and doubt, for reminding me of the big picture when I needed it, and for teaching me life lessons I will not forget! The work presented here would not exist without the help and support of many collaborators. Special thanks to Dr Rina Ariga, for providing us with the rich dataset used in this work, and for all her hard work and patience. I would also like to thank Professor Hugh Watkins for his implication in this research, Dr Ernesto Zacur, for all the work and guidance on the imaging work, Professor Nando de Freitas, for his fruitful ideas, and Professor Pablo Laguna. Pablo, muchas gracias por todas las discusiones tan productivas que hemos tenido, tus geniales ideas y sobre todo por tu apoyo y generosidad. Thank you to the Department of Computer Science, for a fantastic work environment, and the British Heart Foundation for funding my DPhil. A big thank you to all my friends who made these three years amazing: Laure, for being such a good friend, and the best food and sport mate ever, Héctor, for all your inappropriate jokes and your continuous patience with me, Cristian, for your never-ending Italian energy, Anna, my DPhil mum, Nicolas, my best tennis partner, and everyone in the CCS group for making everyday life in the office so nice. Thanks to my friends back in France: Emilie, Adrien, Tatiana, Nils, for staying in touch despite my rare trips back to Paris. Thank you to St Hugh’s College and its MCR for making me feel part of the family, and to all the sporting teams I was lucky to join, especially OULTC, and OULGC. Separately, thank you to Kayla Friedman and Malcolm Morgan of the Centre for Sustainable Development, University of Cambridge, for the Microsoft Word thesis template used to produce this document. Finally, I would like to thank all my family, especially my mum, dad and sister for supporting me every step of the way! vi CONTENTS PUBLICATIONS .............................................................................................................. IV CONFERENCE PRESENTATIONS ....................................................................................... V PRIZES ........................................................................................................................... V 1 INTRODUCTION ....................................................................................................... 1 1.1 INTRODUCTION ........................................................................................................ 1 1.2 THESIS AIMS ............................................................................................................. 4 1.3 THESIS OUTLINE ....................................................................................................... 5 2 CARDIAC ELECTROPHYSIOLOGY AND HYPERTROPHIC CARDIOMYOPATHY .................................................................................................. 8 2.1 INTRODUCTION ........................................................................................................ 8 2.2 ELECTROPHYSIOLOGY OF THE HEALTHY HEART ....................................................... 9 2.2.1 Cardiac anatomy and function ............................................................................. 9 2.2.2 Cardiac electrical propagation: from single cell to whole organ ...................... 11 2.2.3 Electrocardiogram recording ............................................................................. 15 2.3 HYPERTROPHIC CARDIOMYOPATHY (HCM) ........................................................... 19 2.3.1 HCM structural and electrophysiological abnormalities ................................... 20 2.3.2 Risk stratification in HCM .................................................................................. 23 2.3.3 ECG abnormalities and standard biomarkers in HCM ...................................... 25 2.3.4 Imaging techniques for the identification of HCM structural abnormalities ..... 26 3 COMPUTATIONAL METHODS FOR ELECTROCARDIOGRAM ANALYSIS AND INTERPRETATION .......................................................................................... 28 3.1 INTRODUCTION ...................................................................................................... 28 vii 3.2 SIGNAL PROCESSING TECHNIQUES FOR ECG SIGNALS ............................................ 29 3.2.1 Noise removal and filtering of Holter recordings ............................................... 29 3.2.2 Delineation of ECG waveforms .......................................................................... 31 3.3 MACHINE LEARNING TECHNIQUES FOR CLASSIFICATION OF ECG SIGNALS ............ 38 3.3.1 Objectives of state of the art methods and available databases ......................... 38 3.3.2 Feature extraction and dimensionality reduction ............................................... 40 3.3.3 Main machine learning techniques for ECG classification ................................ 43 3.3.4 Discussion ........................................................................................................... 54 3.4 PERSONALIZED WHOLE-BODY COMPUTER SIMULATIONS ........................................ 56 3.4.1 Modelling of cardiac anatomy ............................................................................ 57 3.4.2 Modelling of the cardiac electrical propagation ................................................ 59 3.4.3 Modelling of cardiac heterogeneities ................................................................. 62 4 EXTRACTION OF ECG PHENOTYPES IN HCM BASED ON QRS MORPHOLOGICAL BIOMARKERS ...................................................................... 64 4.1 INTRODUCTION ....................................................................................................... 64 4.2 MOTIVATION .......................................................................................................... 65 4.3 METHODS ............................................................................................................... 66 4.3.1 Clinical database ................................................................................................ 66 4.3.2 ECG signal processing ....................................................................................... 68 4.3.3 Extraction of standard and Hermite based biomarkers ...................................... 71 4.3.4 Binary classification between HCM and controls .............................................. 74 4.3.5 Dimensionality reduction and clustering ............................................................ 75 4.3.6 Statistical analysis .............................................................................................. 77 viii
Description: