Uso da inteligência artificial aplicada ao Eletrocardiograma para diagnóstico de Disfunção Sistólica Ventricular Esquerda

• Ano de defesa: 2023
• Autor(a) principal: Wilton Batista de Santana Júnior
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Minas Gerais; Brasil; MED - DEPARTAMENTO DE CLÍNICA MÉDICA; Programa de Pós-Graduação em Ciências da Saúde - Infectologia e Medicina Tropical; UFMG
• Palavras-chave em Português: inteligência artificial; insuficiência cardíaca; eletrocardiografia; disfunção ventricular esquerda; Inteligência Artificial; Insuficiência Cardíaca; Eletrocardiografia; Disfunção Ventricular Esquerda
• Link de acesso: http://hdl.handle.net/1843/61498

INTRODUCTION: Heart failure (HF) is one of the three most common causes of cardiovascular diseases (CVD), which are the leading causes of morbidity and mortality worldwide. The electrocardiogram (ECG) is one of the tests used in the evaluation of HF, combining low-cost and wide accessibility. When combined with artificial intelligence, the ECG can be a powerful tool for screening individuals with a higher risk of HF. Our objective was to assess the performance of an AI algorithm applied to the ECG for the detection of left ventricular systolic dysfunction (LVSD) and compare it to the performance of major ECG abnormalities (MEA) according to the Minnesota code. METHODS: This was a retrospective cross-sectional diagnostic accuracy study using data from the Brazilian Longitudinal Study of Adult Health (ELSA-Brazil). A total of 2567 individuals with valid echocardiograms (ECO) and ECGs and probability values for left ventricular systolic dysfunction (LVSD) estimated by an artificial intelligence (AI) algorithm, were evaluated. LVSD was defined as a left ventricular ejection fraction (LVEF) less than 40%, calculated using ECO. The prevalence of LVSD was 1.13% in the studied population (29 individuals). Sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), positive likelihood ratio (PLR), negative likelihood ratio (NLR), and diagnostic odds ratio (DOR) were calculated for the algorithm and MEA. The area under the ROC curve (AUC-ROC) was also calculated for the algorithm. RESULTS: The study population had a median age of 62 years, with 47.2% being male. The AUC-ROC for the algorithm to predict HF was 0.947 (95% CI 0.913 – 0.981). Sensitivity, specificity, PPV, NPV, PLR, NLR, and DOR for the algorithm were 0.690, 0.976, 0.244, 0.996, 27.6, 0.32, and 88.74, respectively. For MEA, it was 0.172, 0.837, 0.012, 0.989, 1.09, 0.990, and 1.07, respectively. CONCLUSIONS: AI applied to the ECG is a promising tool for identifying patients with a higher likelihood of HF who should be prioritized for ECO. This could improve the diagnosis capacity of HF in our setting and thus enable early treatment initiation, with possible impact on reducing morbidity and mortality.