Automated pipeline for non-invasive classification and location of atrial arrhythmia mechanisms based on and body surface mapping and electrocardiographic imaging
Introduction: The heart is a vital organ that acts as an electrical-mechanical pump, which transports essential substances to different tissues and organs in the body. Any impairment in its ability to effectively pump blood can lead to dysfunction and irreversible organ damage, even death. Factors like genetics, aging, poor diet, sedentary lifestyle, and excessive alcohol consumption can affect heart health and contribute to heart diseases, including cardiac arrhythmias. Arrhythmias, such as atrial tachycardia (AT), flutter (AFL) and fibrillation (AF), have a high prevalence, been 1-2% for AF in adult population, increasing the risk of thromboembolism, heart failure and stroke. Electrophysiological examinations are typically used to identify arrhythmic regions in the heart but are usually invasive, time-consuming, and costly. Body surface potential mapping (BSPM) is a non-invasive alternative to obtain spatiotemporal information of the electrophysiologic behavior of the heart. The Electrocardiographic imaging (ECGi) is a technic that allows the non-invasive reconstruction of the epicardial signals by using the BSPM signals and the torso and heart geometries, extracted from medical imaging. From this, the present work aims to develop maps and classifiers, based on BSPM and ECGi, that could help the clinicians by providing information about the nature of the arrhythmia and the location of its maintaining mechanism.
Hypothesis: The non-invasive analysis of high-density mapping, such as the BSPM, can provide useful information for the clinicians, aiding their decision making and easing the invasive procedure, consequently reducing the intervention time.
General Objective: Study and develop pipelines to better understand atrial arrhythmias, with possible applications in clinics by given information about the nature of the arrhythmia and the location of its maintaining mechanism.
Specific objectives: Extract biomarkers from BSPM signals and rank them by its contribution in the discrimination of the kind of arrhythmia and the location of the mechanism; Construct classifiers based on the best ranked biomarkers; and Create maps based on the optical flow of reconstructed atrial signals, and locate the maintaining arrhythmic mechanism with these maps.
Methods: This work is divided in two parts: the first part consists in the extraction of biomarkers from BSPM and ECGi, the ranking of the biomarkers based in three metrics, and the creation of classifiers based in logistic regression, using the best ranked classifiers. These classifiers aim to discriminate different arrhythmia mechanism and find its location between left and right atria. The second part of this work consists in a creation of maps based in the optical flow of the reconstructed ECGi signals, and its use in the location of functional rotors (FR) and ectopic foci (EF).
This work used a dataset composed of 22 signals from realistic in silico models of the atria in arrhythmic conditions (AF=10, AT=8, AFL=4), composed of 284,578 nodes and 1,353,783 tetrahedrons.
In the first part of the work, 40 biomarkers were ranked, been extracted from Phase, Dominant Frequency, Organization Index, Local Activation Time and Optical Flow maps. The pipeline ranked the biomarker’s contribution in the classification of the maintaining arrhythmic mechanism and its location, basing on a combination of three different indexes: Analysis of variation (ANOVA), Kendall tau and Lasso. The two best biomarkers for each arrythmia classification were used in a logistic-regression-based algorithm. For the classification of the mechanism location, the same biomarkers were extracted from 4 different areas of the maps (anterior posterior, left, and right) and ranked based in its contribution to the spatial location (left or right atrium) of the mechanism or circuit.
In the second part of the work, the reconstructed signals of the atrial, obtained through ECGi, was used in the Farnebäck Optical Flow (FO) algorithm, by using an element-wise projection, interpolation and bidimensional processing. That is, all processing was computed for each vertex of the atria geometry. The processing consists in the obtention of the motion vector field (MVF) with FOF, and the computation of the curl and the divergent of this MVF, generating two maps, curl map (CM) and divergent map (DM). Then the value obtained in the center of the CM and the DM computed for each vertex was assigned to this vertex, generating a three-dimensional CM (CM3D) and DM (DM3D). The vertex with maximum absolute value in the CM3D was considered a FR, and the maximum value of the DM3D was considered an EF. The accuracy of the mechanism location was calculated by comparing the results for the reconstructed atrial signals and the original signals, and considering the location within a normalized error of 10% as a true positive. Also, was calculated the accuracy of detecting the mechanism in the correct atrium and the mean normalized error between the original and located positions.
Results: For the classification section, the mechanism classification obtained an overall accuracy of 0.864, with the location accuracy of 0.818, on the torso maps, and on epicardium maps the accuracy is yet to be calculated. The novel Optical Flow (along with phase and frequency) related biomarkers showed a great contribution in the location and classification of the arrhythmias’ mechanisms.
For the second section, the proposed method could locate the mechanisms with a mean normalized error of 7% for the EF mechanism, and 19% for FR. Different values of signal to noise ratio (SNR), did not seem to affect the accuracy.
Conclusions: Noninvasive maps allow extraction of biomarkers to classify arrhythmia and locate its mechanism in the atria non-invasively. Also, the analysis of the spatial dynamic of the signals with optical flow applied to reconstructed atrial signals showed a great contribution to the location of arrhythmic mechanisms even before the invasive intervention. Analysis with BSPM and ECGi seems to have a great potential in aiding the treatment and diagnosis of atrial arrhythmias.