Universidade Federal do Ceará
Mestrado e Doutorado em Ciências da Computação

Título: Anomaly Detection in Electrocardiogram Signals Using Advanced Unsupervised Machine Learning for Arrhythmia Identification

Data: 25/11/2024
Horário: 16h00
Local: Sala de Seminários – Bloco 952

Resumo:

Anomaly detection in Electrocardiogram (ECG) data is critical for the early diagnosis and treatment of cardiac conditions, especially in identifying arrhythmias. This study employs unsupervised machine learning and deep learning techniques to detect anomalies in ECG signals, focusing on feature engineering and direct time-series analysis through recurrent neural networks. Our methodology combines two primary approaches: (i) structured feature extraction from ECG examinations, facilitating traditional anomaly detection models, and (ii) feature learning directly from raw ECG signals using a Long Short-Term Memory (LSTM) Autoencoder, designed to capture normal temporal patterns in ECG data, enabling the identification of cardiac anomalies as deviations from these patterns. Data from the MIT-BIH Arrhythmia Database was used to evaluate model performance, demonstrating that the LSTM Autoencoder achieved the highest AUC-ROC score, indicating superior sensitivity and specificity in detecting arrhythmic events. The results highlight the potential of unsupervised learning for automatic feature extraction and real-time anomaly detection, supporting the development of AI-based diagnostic tools for wearable devices that provide continuous cardiovascular monitoring and early intervention capabilities.

Banca examinadora: