Classificação de fibrilação atrial em ECG com métricas de convergência da decomposição tensorial BTD

Abstract

Amid the growing incidence of cardiovascular diseases, arrhythmias play a central role in global morbidity and mortality. Among them, atrial fibrillation stands out as the most common sustained arrhythmia and is associated with a high risk of complications such as stroke and heart failure. Given this scenario, it is essential to develop tools that help in the early and accurate diagnosis of this condition. In this context, this work investigates and expands the applicability of convergence metrics derived from the tensor decomposition BTD in ECG signals for the classification of atrial fibrillation. In a context in which diagnostic accuracy and speed are imperative, the purpose of this study is to extract attributes from the BTD convergence indicators and combine them with various machine learning models. The methodology adopted involved the comparative analysis of various combinations of attributes and the implementation of models ranging from traditional approaches to advanced architectures based on RNN. The experimental results showed that both the conventional models and the models based on RNN, when using the extracted attributes, achieved accuracies of around 95% and F1-Score of 96.50%, corroborating the robustness and effectiveness of the proposed indicators for distinguishing between atrial fibrillation and normal sinus rhythm. In addition, the study highlights the importance of integrating tensor decomposition techniques and machine learning algorithms to improve diagnostic methods in cardiology. Finally, it is concluded that the findings presented encourage future research aimed at optimizing attribute vectors and refining the parameters of deep learning models, as well as evaluating the computational efficiency of the proposed approaches, contributing significantly to the advancement of automated diagnosis of cardiac conditions.