Advanced search
1 file | 5.21 MB Add to list

Equine electrocardiography revisited : 12- lead recording, vectorcardiography and the power of machine intelligence

(2020)
Author
Promoter
(UGent) , (UGent) , (UGent) and Tammo Delhaas
Organization
Abstract
Arrhythmias are very common in horses, but the origin and clinical significance is often unknown. The use of the electrocardiogram (ECG) in horses is still limited to rhythm and rate diagnosis whereas identification of the underlying mechanisms of arrhythmias often requires invasive electrophysiological studies that have not fully been developed in horses. Meanwhile, in human medicine the 12-lead ECG has been a cornerstone in the development of human cardiology. Next to the diagnosis of arrhythmias, it also allows recognition of structural cardiac abnormalities, conduction disturbances, genetically mediated electrical abnormalities, myocardial infarction and electrolyte disturbances. Additionally, advanced invasive techniques have been developed in humans for both the identification of the underlying mechanisms of arrhythmias and treatment of arrhythmias. Equine cardiology has not followed this evolution largely because knowledge from human and small animal cardiology cannot be directly extrapolated to horses due to their different conduction system. For a long time, it was believed that only a part of the depolarization of the equine heart was visible on the surface ECG and that, hence, the surface ECG was only of limited use. The General introduction discusses the current knowledge about human and especially equine ECG. First the genesis and nomenclature of the electrocardiogram in horses is briefly discussed. Next, an overview is given of the development of ECG in humans with extra emphasis on the 12-lead ECG and vectorcardiography (VCG). Subsequently, the anatomical and physiological differences between the human and equine heart are discussed. These are followed by the development of equine ECG and some common lead systems that have been used in horses. Subsequently, equine ECG characteristics and analysis are described. This section starts with the description of the most commonly occurring arrhythmias, their underlying pathophysiology and their appearance on the surface ECG. Thereafter, manual analyses of the equine ECG are described followed by a description of automated analyses of the human ECG. Finally, a short overview is given of invasive cardiac electrophysiological techniques in humans and horses. The first major objective of this dissertation was to improve the screening and diagnosis of arrhythmias in horses using the modified base-apex bipolar surface ECG recording. Therefore, we explored the use of deep learning for the semi-automated analysis of equine ECGs in Chapter 1 and evaluated the use of the atrial fibrillatory rate (AFR) derived from a surface ECG in Chapter 2. The second major objective was to develop a new invasive method for identifying the origin of cardiac arrhythmias: 3D electro-anatomical mapping. Therefore, the development and methodology of the technique in horses is described in Chapter 3, and the normal depolarization pattern of the heart in sinus rhythm (SR) using this 3D electro-anatomical mapping technique is described in Chapter 4. The third major objective of this study was to evaluate if multiple lead recordings, i.e. the 12-lead ECG and vectorcardiography, have additional value for equine cardiology if adapted to equine physiology. Therefore, we explored the use of 12-lead ECG and VCG for the determination of the site of origin for atrial and ventricular premature depolarizations in Chapter 5. Chapter 1 describes the development of a complete algorithm for equine specific ECG analysis. Both the initial filtering and QRS beat detection were done with wavelet transformations. The QRS detection algorithm outperformed the classic Pan-Tompkins algorithm with a sensitivity of 99.0% versus 91.5% for the Pan-Tompkins algorithm. Next, a novel parallel convolutional neural network architecture was proposed for the feature extraction and classification of the individual beats. The novelty of this network architecture is the parallel processing of both the morphological data from the ECG deflections and the relative timing to the other beats in the ECG. Because no public datasets are available for equine ECGs, a dataset was made with 26.440 beats in 4 classes: normal, ventricular and atrial premature depolarization and noise. The network was then trained and tested using both the human MIT-BIH arrhythmia dataset and the own-made equine ECG dataset. The network was optimized using a genetic algorithm and an accuracy of 97.7% and 92.6% was achieved for the MIT-BIH and equine ECG database, respectively. Afterwards, transfer learning from the MIT-BIH dataset to the equine ECG database was applied after which the average accuracy, sensitivity, positive predictive value and F1 score of the network increased with an accuracy of 97.1%. In Chapter 2 the AFR derived from a single-lead surface ECG was compared with the atrial fibrillation rate derived from a right atrial intracardiac recording (RA-FR). Three minute long episodes of simultaneous electrograms and surface ECG during atrial fibrillation (AF) were selected for analysis and compared using Bland-Altman test. The mean RA-FR was measured from the deflections on the intracardiac electrogram, while the AFR was extracted from the surface ECG using spatiotemporal QRS and T-wave cancellation. In addition, we evaluated the correlation of AFR with transvenous electrical cardioversion (TVEC) threshold (in Joules), number of shocks and cardioversion success rate in horses. This was done in 73 horses treated for atrial fibrillation by TVEC. The mean difference between RA-FR and AFR was -13 fibrillations per minute and there was a moderate (r=0.65) correlation between RA-FR and AFR. Neither RA-FR nor AFR showed a significant correlation with transvenous electrical cardioversion threshold, number of shocks or cardioversion success. We concluded that the AFR may allow non-invasive long-term monitoring of AF dynamics and that neither RA-FR nor AFR can be used to predict the minimal defibrillation threshold. The methodology for 3D ultra-high-density electro-anatomical mapping is described in Chapter 3. Ultra-high-density cardiac mapping allowed very accurate characterization of atrial and ventricular electrophysiology and activation timing. Electro-anatomical maps were acquired from all 4 chambers in four horses in SR under general anaesthesia. All endocardial areas within each chamber could be reached with the basket catheter of the mapping system, but access to the left atrium required the use of a deflectable sheath. With the exception of the left atrial map of 1 horse, all four chambers in all four horses could be mapped. The mapping system uses beat acceptance criteria for the automatic acquisition of endocardial electrograms. Optimization of the beat acceptance criteria led to a reduction in manual correction of the automatically accepted beats from 13.1% in the first horse to 0.4% of the beats in the last horse. The study shows that 3D electro-anatomical mapping is feasible in adult horses and that it is a promising tool for electrophysiological research and characterization of complex arrhythmias. The first electrophysiological research performed with the 3D electro-anatomical mapping system in horses is described in Chapter 4. The system was used to evaluate the qualitative and quantitative depolarization patterns and correlation to the surface ECG of both the atrial and ventricular endocardium in 7 healthy horses in sinus rhythm under general anaesthesia. This was done by analysing the bipolar activation maps of the endocardium. The first atrial activation was located at the height of the terminal crest. Only one interatrial conduction pathway was recognized. The first and second P wave deflections represented the right and left atrial depolarization, respectively. Bundle of His electrograms could be recorded in 5 out of 7 horses. Left ventricular activation started at the mid septum and right ventricular activation started apically from the supraventricular crest. This was followed by separate depolarizations at the height of the right and left ventricular mid free wall. Further ventricular depolarization occurred in an explosive pattern. The results of this study are a reference for the normal sinus impulse propagation pattern and for the conduction velocities in equine atria and ventricles. Even more importantly, these results show that all parts of the atrial and ventricular depolarization contribute to the surface ECG’s P wave and QRS complex. Thus, the surface ECG does contain information about the entire depolarization of the heart, in contrast to findings of previous studies. The next goal was to evaluate whether VCG characteristics can differentiate between anatomical locations of atrial (APDs) and ventricular premature depolarizations (VPDs) as well as between SR and APDs (Chapter 5). In 7 horses a 12-lead ECG was recorded under general anaesthesia while endomyocardial pacing was performed (800-1000 ms cycle length) in the atria at the pulmonary veins, left atrial free wall and septum, right atrial free wall, intervenous tubercle, as well as the cranial and caudal junction with vena cava. Endomyocardial pacing was performed in the ventricles at the apex, mid and high septum and mid and high free wall, and at the right ventricular outflow tract. Catheter positioning was guided by 3D electro-anatomical mapping and transthoracic ultrasound. The VCG was calculated from the 12-lead ECG using custom-made algorithms and was used to determine the mean electrical axis of the first and second half of the P wave and the initial and mean electrical axis of the QRS complex. A significant differentiation could be made between the site of origin of every paced APD. SR could be differentiated from all paced APDs except those originating from the cranial junction with vena cava. For the ventricles a significant differentiation was possible between left and right ventricular paced complexes. Within the left ventricle, paced complexes from all locations showed significant differences. Within the right ventricle, only paced complexes originating from the right ventricular outflow tract were significantly different from all other paced beats. Paced complexes originating from other locations within the right ventricle only showed significant differences compared to locations not adjacent to the paced location. Paced complexes originating from the right ventricular mid septum and the left ventricular apex were not significantly different from sinus rhythm. These results suggest that VCG, and in extension multiple lead recordings, could be useful to identify the anatomical origin of atrial and ventricular ectopy in horses. However, differentiation of the site of origin within the right ventricle is challenging. As a general conclusion this work shows that equine ECGs have more clinical applications than currently employed and especially multiple lead recordings are underused at the moment. The results and methods described in this thesis can be a starting point for future investigations into equine electrocardiography. The algorithm for semi-automated analysis of equine ECGs could be used to improve screening for arrythmias in the general horse population. The atrial fibrillation rate derived from the surface ECG allows to better understand the underlying pathophysiology of atrial fibrillation. Finally, 3D electro-anatomical mapping and multiple lead recordings can be used to determine the underlying origin of arrhythmias and guide their targeted treatment.
Keywords
Cardiology, equine, arrhythmias, deep learning, 3D electro-anatomical mapping, electrocardiography, vectorcardiography

Downloads

  • (...).pdf
    • full text (Published version)
    • |
    • UGent only
    • |
    • PDF
    • |
    • 5.21 MB

Citation

Please use this url to cite or link to this publication:

MLA
Van Steenkiste, Glenn. Equine Electrocardiography Revisited : 12- Lead Recording, Vectorcardiography and the Power of Machine Intelligence. Universiteit Gent. Faculteit Diergeneeskunde, 2020.
APA
Van Steenkiste, G. (2020). Equine electrocardiography revisited : 12- lead recording, vectorcardiography and the power of machine intelligence. Universiteit Gent. Faculteit Diergeneeskunde.
Chicago author-date
Van Steenkiste, Glenn. 2020. “Equine Electrocardiography Revisited : 12- Lead Recording, Vectorcardiography and the Power of Machine Intelligence.” Universiteit Gent. Faculteit Diergeneeskunde.
Chicago author-date (all authors)
Van Steenkiste, Glenn. 2020. “Equine Electrocardiography Revisited : 12- Lead Recording, Vectorcardiography and the Power of Machine Intelligence.” Universiteit Gent. Faculteit Diergeneeskunde.
Vancouver
1.
Van Steenkiste G. Equine electrocardiography revisited : 12- lead recording, vectorcardiography and the power of machine intelligence. Universiteit Gent. Faculteit Diergeneeskunde; 2020.
IEEE
[1]
G. Van Steenkiste, “Equine electrocardiography revisited : 12- lead recording, vectorcardiography and the power of machine intelligence,” Universiteit Gent. Faculteit Diergeneeskunde, 2020.
@phdthesis{8667065,
  abstract     = {Arrhythmias are very common in horses, but the origin and clinical significance is often unknown. The use of the electrocardiogram (ECG) in horses is still limited to rhythm and rate diagnosis whereas identification of the underlying mechanisms of arrhythmias often requires invasive electrophysiological studies that have not fully been developed in horses. Meanwhile, in human medicine the 12-lead ECG has been a cornerstone in the development of human cardiology. Next to the diagnosis of arrhythmias, it also allows recognition of structural cardiac abnormalities, conduction disturbances, genetically mediated electrical abnormalities, myocardial infarction and electrolyte disturbances. Additionally, advanced invasive techniques have been developed in humans for both the identification of the underlying mechanisms of arrhythmias and treatment of arrhythmias. Equine cardiology has not followed this evolution largely because knowledge from human and small animal cardiology cannot be directly extrapolated to horses due to their different conduction system. For a long time, it was believed that only a part of the depolarization of the equine heart was visible on the surface ECG and that, hence, the surface ECG was only of limited use. 
The General introduction discusses the current knowledge about human and especially equine ECG. First the genesis and nomenclature of the electrocardiogram in horses is briefly discussed. Next, an overview is given of the development of ECG in humans with extra emphasis on the 12-lead ECG and vectorcardiography (VCG). Subsequently, the anatomical and physiological differences between the human and equine heart are discussed. These are followed by the development of equine ECG and some common lead systems that have been used in horses. Subsequently, equine ECG characteristics and analysis are described. This section starts with the description of the most commonly occurring arrhythmias, their underlying pathophysiology and their appearance on the surface ECG. Thereafter, manual analyses of the equine ECG are described followed by a description of automated analyses of the human ECG. Finally, a short overview is given of invasive cardiac electrophysiological techniques in humans and horses.
The first major objective of this dissertation was to improve the screening and diagnosis of arrhythmias in horses using the modified base-apex bipolar surface ECG recording. Therefore, we explored the use of deep learning for the semi-automated analysis of equine ECGs in Chapter 1 and evaluated the use of the atrial fibrillatory rate (AFR) derived from a surface ECG in Chapter 2. The second major objective was to develop a new invasive method for identifying the origin of cardiac arrhythmias: 3D electro-anatomical mapping. Therefore, the development and methodology of the technique in horses is described in Chapter 3, and the normal depolarization pattern of the heart in sinus rhythm (SR) using this 3D electro-anatomical mapping technique is described in Chapter 4. The third major objective of this study was to evaluate if multiple lead recordings, i.e. the 12-lead ECG and vectorcardiography, have additional value for equine cardiology if adapted to equine physiology. Therefore, we explored the use of 12-lead ECG and VCG for the determination of the site of origin for atrial and ventricular premature depolarizations in Chapter 5.
Chapter 1 describes the development of a complete algorithm for equine specific ECG analysis. Both the initial filtering and QRS beat detection were done with wavelet transformations. The QRS detection algorithm outperformed the classic Pan-Tompkins algorithm with a sensitivity of 99.0% versus 91.5% for the Pan-Tompkins algorithm. Next, a novel parallel convolutional neural network architecture was proposed for the feature extraction and classification of the individual beats. The novelty of this network architecture is the parallel processing of both the morphological data from the ECG deflections and the relative timing to the other beats in the ECG. Because no public datasets are available for equine ECGs, a dataset was made with 26.440 beats in 4 classes: normal, ventricular and atrial premature depolarization and noise. The network was then trained and tested using both the human MIT-BIH arrhythmia dataset and the own-made equine ECG dataset. The network was optimized using a genetic algorithm and an accuracy of 97.7% and 92.6% was achieved for the MIT-BIH and equine ECG database, respectively. Afterwards, transfer learning from the MIT-BIH dataset to the equine ECG database was applied after which the average accuracy, sensitivity, positive predictive value and F1 score of the network increased with an accuracy of 97.1%.
In Chapter 2 the AFR derived from a single-lead surface ECG was compared with the atrial fibrillation rate derived from a right atrial intracardiac recording (RA-FR). Three minute long episodes of simultaneous electrograms and surface ECG during atrial fibrillation (AF) were selected for analysis and compared using Bland-Altman test. The mean RA-FR was measured from the deflections on the intracardiac electrogram, while the AFR was extracted from the surface ECG using spatiotemporal QRS and T-wave cancellation. In addition, we evaluated the correlation of AFR with transvenous electrical cardioversion (TVEC) threshold (in Joules), number of shocks and cardioversion success rate in horses. This was done in 73 horses treated for atrial fibrillation by TVEC. The mean difference between RA-FR and AFR was -13 fibrillations per minute and there was a moderate (r=0.65) correlation between RA-FR and AFR. Neither RA-FR nor AFR showed a significant correlation with transvenous electrical cardioversion threshold, number of shocks or cardioversion success. We concluded that the AFR may allow non-invasive long-term monitoring of AF dynamics and that neither RA-FR nor AFR can be used to predict the minimal defibrillation threshold.
The methodology for 3D ultra-high-density electro-anatomical mapping is described in Chapter 3. Ultra-high-density cardiac mapping allowed very accurate characterization of atrial and ventricular electrophysiology and activation timing. Electro-anatomical maps were acquired from all 4 chambers in four horses in SR under general anaesthesia. All endocardial areas within each chamber could be reached with the basket catheter of the mapping system, but access to the left atrium required the use of a deflectable sheath. With the exception of the left atrial map of 1 horse, all four chambers in all four horses could be mapped. The mapping system uses beat acceptance criteria for the automatic acquisition of endocardial electrograms. Optimization of the beat acceptance criteria led to a reduction in manual correction of the automatically accepted beats from 13.1% in the first horse to 0.4% of the beats in the last horse. The study shows that 3D electro-anatomical mapping is feasible in adult horses and that it is a promising tool for electrophysiological research and characterization of complex arrhythmias.
The first electrophysiological research performed with the 3D electro-anatomical mapping system in horses is described in Chapter 4. The system was used to evaluate the qualitative and quantitative depolarization patterns and correlation to the surface ECG of both the atrial and ventricular endocardium in 7 healthy horses in sinus rhythm under general anaesthesia. This was done by analysing the bipolar activation maps of the endocardium. The first atrial activation was located at the height of the terminal crest. Only one interatrial conduction pathway was recognized. The first and second P wave deflections represented the right and left atrial depolarization, respectively. Bundle of His electrograms could be recorded in 5 out of 7 horses. Left ventricular activation started at the mid septum and right ventricular activation started apically from the supraventricular crest. This was followed by separate depolarizations at the height of the right and left ventricular mid free wall. Further ventricular depolarization occurred in an explosive pattern. The results of this study are a reference for the normal sinus impulse propagation pattern and for the conduction velocities in equine atria and ventricles. Even more importantly, these results show that all parts of the atrial and ventricular depolarization contribute to the surface ECG’s P wave and QRS complex. Thus, the surface ECG does contain information about the entire depolarization of the heart, in contrast to findings of previous studies.
The next goal was to evaluate whether VCG characteristics can differentiate between anatomical locations of atrial (APDs) and ventricular premature depolarizations (VPDs) as well as between SR and APDs (Chapter 5). In 7 horses a 12-lead ECG was recorded under general anaesthesia while endomyocardial pacing was performed (800-1000 ms cycle length) in the atria at the pulmonary veins, left atrial free wall and septum, right atrial free wall, intervenous tubercle, as well as the cranial and caudal junction with vena cava. Endomyocardial pacing was performed in the ventricles at the apex, mid and high septum and mid and high free wall, and at the right ventricular outflow tract. Catheter positioning was guided by 3D electro-anatomical mapping and transthoracic ultrasound. The VCG was calculated from the 12-lead ECG using custom-made algorithms and was used to determine the mean electrical axis of the first and second half of the P wave and the initial and mean electrical axis of the QRS complex. A significant differentiation could be made between the site of origin of every paced APD. SR could be differentiated from all paced APDs except those originating from the cranial junction with vena cava. For the ventricles a significant differentiation was possible between left and right ventricular paced complexes. Within the left ventricle, paced complexes from all locations showed significant differences. Within the right ventricle, only paced complexes originating from the right ventricular outflow tract were significantly different from all other paced beats. Paced complexes originating from other locations within the right ventricle only showed significant differences compared to locations not adjacent to the paced location. Paced complexes originating from the right ventricular mid septum and the left ventricular apex were not significantly different from sinus rhythm. These results suggest that VCG, and in extension multiple lead recordings, could be useful to identify the anatomical origin of atrial and ventricular ectopy in horses. However, differentiation of the site of origin within the right ventricle is challenging.
As a general conclusion this work shows that equine ECGs have more clinical applications than currently employed and especially multiple lead recordings are underused at the moment. The results and methods described in this thesis can be a starting point for future investigations into equine electrocardiography. The algorithm for semi-automated analysis of equine ECGs could be used to improve screening for arrythmias in the general horse population. The atrial fibrillation rate derived from the surface ECG allows to better understand the underlying pathophysiology of atrial fibrillation. Finally, 3D electro-anatomical mapping and multiple lead recordings can be used to determine the underlying origin of arrhythmias and guide their targeted treatment.},
  author       = {Van Steenkiste, Glenn},
  keywords     = {Cardiology,equine,arrhythmias,deep learning,3D electro-anatomical mapping,electrocardiography,vectorcardiography},
  language     = {eng},
  pages        = {213},
  publisher    = {Universiteit Gent. Faculteit Diergeneeskunde},
  school       = {Ghent University},
  title        = {Equine electrocardiography revisited : 12- lead recording, vectorcardiography and the power of machine intelligence},
  year         = {2020},
}