Desenvolvimento e validação de detector de pico R baseado em vetorcardiograma

• Ano de defesa: 2019
• Autor(a) principal: Thiago Lucas de Oliveira
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Minas Gerais; Brasil; ENG - DEPARTAMENTO DE ENGENHARIA ELÉTRICA; Programa de Pós-Graduação em Engenharia Elétrica; UFMG
• Palavras-chave em Português: ECG; Detecção de pico R; Vetorcardiograma; Métricas de desempenho; Regra de múltiplas derivações; Engenharia elétrica; Engenharia biomédica; Eletrocardiograma
• Link de acesso: http://hdl.handle.net/1843/31880

The electrocardiogram (ECG) is considered one of the most important diagnosis assistance methods in the cardiovascular clinical area. With the expansion of the ECG exam acquisition, it has become necessary to develop algorithms for automating the ECG signal analysis, reducing the medical report time. In this context, the objective of this research is the development of an algorithm for automatic detection of R peaks on short 12-leads ECG excerpt based on vectorcardiogram. This algorithm, named Vectordet, considers the health systems limitations, which are the excessive noise level, diverse pathologies and limited computational resources. This case scenario is represented by St. Petersburg Institute of Cardiological Technics 12-lead Arrhythmia Database. The Vectordet performance is compared with the references Pan-Tompkins and Wavedet, which are single lead R peak detection algorithms. Before the distinct capacity in each lead detection of these two references, a multiple leads rule, called Single Lead Rule, was studied for weighting different input detections. The algorithms performances evaluation is restricted to the metrics of sensitivity (Se), positive predictivity (P+) and processing time (Tp), although it is presented the RMS error, specificity and accuracy. Thereby, the results obtained in the proposed algorithm are Se = 99.30%, P+ = 99.15% and a Tp = 0.001 s for each 10-s exam. From the reference algorithms, Vectordet ensures an R peak detection with a performance statistically superior to Wavedet and a computational cost 25 and 3700 times faster than Pan-Tompkins and Wavedet, respectively.