Banca de DEFESA: VICTOR GONÇALVES MARQUES

Uma banca de DEFESA de MESTRADO foi cadastrada pelo programa.
DISCENTE : VICTOR GONÇALVES MARQUES
DATA : 15/05/2020
HORA: 10:00
LOCAL: Bloco Delta - São Bernardo do Campo
TÍTULO:

Characterization of atrial fibrillation in body surface potential mapping systems: a clinical-computational study

PÁGINAS: 90
RESUMO:

Introduction: Atrial fibrillation (AF) is the sustained cardiac arrhythmia most common in clinical practice, affecting between 1 and 2% of the world population. In the last decades, basic and clinical research has made great progress in improving the diagnosis and treatment of AF, and its mechanism has been gradually elucidated, but not fully understood. Invasive electrophysiological studies are the most straightforward approach to identifying regions maintaining AF. However, this strategy is technically complex, demanding a large amount of time, resources and with risks for the patients, motivating the development of non-invasive methods, such as body surface potential mapping (BSPM) and electrocardiographic imaging (ECGi). Among these techniques, BSPM is the simplest approach, demanding relatively simple hardware and software to analyse the signals, and yielding relevant clinical results, which can be of use both in early diagnosis and follow-up, being an important auxiliary tool.

Objective: The main objective of this study is to propose a robust approach for frequency analysis in AF BSPM signals allowing the development of more realistic maps related to the AF pathophysiology and automatically classify this arrhythmia noninvasively to closely related supraventricular atrial arrhythmias (atrial tachycardia - AT, atrial flutter – AFL).

Methods: in this study 567 BSPM leads from realistic 3D computer model of atria arrhythmias (4 ectopic activity - AT, 4 macro-reentrant mechanism - AFL and 11 functional reentry – AF) and 67 BSPM leads from 12 AF patients (9 paroxysmal and 3 persistent) were considered for analysis. The analysis consisted of generating 3D maps of dominant frequency (DF), phase and singularity points. DF maps of the atrial driving frequencies were estimated non-invasively by combining activation detection and DF in Welch periodograms, with a spatial mask to avoid harmonics, and compared with a proposed developed wavelet transform-based method. The robustness of proposed method was tested in different protocols (increasing levels of noise and harmonics presence). Phase maps were obtained applying Hilbert transform on signals filtered around the driver frequency (±1Hz) and the spatiotemporal distribution of phase singularity points (SPs) was analyzed using histograms (heatmaps) and connecting SPs along time (filaments). The impact of reducing the number of leads to realistic clinical scenario set of electrodes (252 to 16) was quantified using similarity measures and statistical tests. Three classifiers (least squares, k-nearest neighbors and random forests) were applied to distinguish automatically the arrhythmias (AT, AFL and AF) based on features extracted from frequency and phase analyses.

Results: analysis of BSPM signals undergoing AF showed that the wavelet-based proposed method outperformed the worldwide Welch approach both in models (correct HDF detection 81.82% vs 45.45%, respectively) and patients (75.00% vs 66.67%). The method was more robust to white Gaussian noise and harmonics and presented more consistent results in lead layouts with low spatial resolution (64 and 32 leads). Moreover, frequency and phase analyses revealed distinct behaviour between AF and other two arrhythmias (AT and AFL). Driver mechanism frequencies were estimated with the highest DF on the torso ignoring 6% of the highest DF values to avoid small harmonic regions, resulting in errors of 12.5 ± 4.8%, 5.21 ± 6.25%, and 9.94 ± 7.16% for AT, AFL and AF, respectively. These frequencies were reflected in a smaller portion of the torso for AF than AT or AFL (p<0.05), having a spatial correlation to the atrial position of the arrhythmic drivers. Filament durations were shorter in AF (p<0.05), followed by AT and AFL. Average rotation frequency from the filaments was similar to the estimated driver frequency. SP clusters in heatmaps were smaller in AFL than AF but had higher density (p<0.01); AT presented intermediate values (p>0.05). No significant differences were found in features extracted from frequency or phase analyses along the different leads’ layouts (252 to 16 leads), despite a loss in sensitivity and precision in SP detection. These features resulted in a automatic classification between AF, AFL and AT with a balanced accuracy of 87.94% using a random forest classifier (40 trees, depth 2).
Conclusion: through a pipeline of signal processing techniques, AF characteristics estimation were improved and more related with the AF pathophysiology. Frequency and phase domains showed intrinsic personalized characteristics across the AT, AFL and AF. The behavior of the arrhythmias could be detected even with low spatial resolution BSPM layouts and were enough for distinguishing automatically the different mechanisms.

MEMBROS DA BANCA:
Presidente - Interno ao Programa - 2231661 - JOAO LOURES SALINET JUNIOR
Membro Titular - Examinador(a) Externo à Instituição - NELSON SAMESIMA
Membro Titular - Examinador(a) Externo à Instituição - OLAF DÖSSEL - KARLSRUHE
Membro Suplente - Examinador(a) Interno ao Programa - 2390463 - PRISCYLA WALESKA TARGINO DE AZEVEDO SIMOES
Membro Suplente - Examinador(a) Interno ao Programa - 2188954 - ERICK DARIO LEON BUENO DE CAMARGO
Membro Suplente - Examinador(a) Externo ao Programa - 2418478 - JOHN ANDREW SIMS