Banca de DEFESA: ANGÉLICA DRIELLY SANTOS DE QUADROS
Uma banca de DEFESA de MESTRADO foi cadastrada pelo programa.STUDENT : ANGÉLICA DRIELLY SANTOS DE QUADROS
Date: 14/11/2025
TIME: 09:00
LOCAL: Sala 107 Bloco Zeta Campus São Bernardo do Campo
TITLE:
Animal model for non-invasive electrocardiographic imaging during cardiac rhythms
PAGES: 80
BIG AREA: Engenharias
AREA: Engenharia Biomédica
SUBÁREA: Bioengenharia
SPECIALTY: Processamento de Sinais Biológicos
SUMMARY:
Electrocardiographic imaging (ECGI) is a promising non-invasive technique that reconstructs epicardial potentials by combining high-density body-surface recordings with patient-specific 3D geometries. An experimental setup and signal processing pipeline was developed to investigate ECGI in isolated Langendorff-perfused rabbit hearts. Panoramic optical mapping, epicardial electrograms, and torso-tank signals were acquired simultaneously under sinus rhythm with induced atrioventricular (AV) block and during tachycardia. After customized preprocessing, the epicardial potentials were reconstructed using Tikhonov (orders 0, 1, 2), Truncated Singular Value Decomposition (TSVD, orders 0, 1, 2), Damped Singular Value Decomposition (DSVD), Generalized Minimal Residual (GMRes), and Bayes regularization methods, with regularization parameter optimized using the L-curve method.
The outcomes showed that no single method was universally optimal, highlighting a trade-off between waveform fidelity, spatiotemporal accuracy, and signal stability. In sinus rhythm, Tikhonov order 1 achieved the best performance in the right atrium (RA) with a mean cross-correlation (CC) of 0.841, while Tikhonov order 2 was superior in the left atrium (LA) with a mean CC of 0.777. During high-rate ventricular tachycardia, Tikhonov order 2 again yielded the highest waveform similarity with a mean CC of 0.828. However, this was accompanied by severe instability, such that the reconstructed amplitudes reached non-physiological levels (3.52 × 105 μV). In contrast, methods like TSVD and DSVD produced realistic amplitudes (14–49 μV), but with slightly lower correlations. Spatial error localization was prominent in tachycardia, with GMRes providing the most accurate result at 7.04 mm. Across all methods, the dominant activation frequencies for both sinus (1.7 Hz) and tachycardia (∼4.7 Hz) were correctly captured.
In parallel, a prototype deep-learning framework, that combines 3D-convolutional encoders with Long Short-Term Memory (LSTM) layers, is under development. Results indicate the potential for enhanced fidelity. Overall, this work demonstrates that the experimental setup delivered reliable recordings and the optimal ECGI strategy is dependent on the rhythm and the region of interest.
COMMITTEE MEMBERS:
Presidente - Interno ao Programa - 2231661 - JOAO LOURES SALINET JUNIOR
Membro Titular - Examinador(a) Externo à Instituição - MARIA DE LA SALUD GUILLEM SANCHEZ - UPV
Membro Titular - Examinador(a) Externo à Instituição - ÓSCAR BARQUERO PÉREZ - URJC
Membro Suplente - Examinador(a) Interno ao Programa - 2188954 - ERICK DARIO LEON BUENO DE CAMARGO
Membro Suplente - Examinador(a) Interno ao Programa - 1188948 - JOAO LAMEU DA SILVA JUNIOR
Membro Suplente - Examinador(a) Externo à Instituição - MATHEUS CARDOSO MORAES - UNIFESP