Prediction of echocardiographic parameters in Chagas disease using heart rate variability and machine learning. (May 2021)

• Record Type: Journal Article
• Authors: Silva, Luiz Eduardo Virgilio; Moreira, Henrique Turin; Bernardo, Mariani Mendes Madisson; Schmidt, André; Romano, Minna Moreira Dias; Salgado, Hélio Cesar; Fazan, Rubens; Tinós, Renato; Marin-Neto, J. Antônio
• Is Part Of: Biomedical signal processing and control. Volume 67(2021)
• Publication Date: 2021-05
• Subjects: Chagas disease -- Heart rate variability -- Machine learning -- Echocardiography -- Risk factor; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2021.102513
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: The prognostic assessment of patients with Chagas Disease (CD) is still a challenge. The most important severity score for CD requires a thorough set of exams. We studied the link between heart rate variability (HRV) and echocardiography in CD. HRV can be used to predict numerical and categorical echocardiographic parameters. HRV is a promising tool to characterize the severity of CD. Abstract: Objective: Investigate whether heart rate variability (HRV) indices can be used to predict morpho-functional parameters obtained from the echocardiogram in a population of patients with Chagas disease (CD). Methods: Sixty-three patients with CD and a recent echocardiogram had their ECG and respiratory signals recorded for 15 min. The cardiac interval series were generated from the ECG and 27 HRV indices, plus the respiratory frequency, were calculated. The correlation between HRV and echocardiographic variables was estimated. The HRV indices were also utilized as inputs in four machine learning schemes to create predictive models for numeric and categorical echocardiographic parameters. Attribute selection schemes were also performed to identify the subset of HRV indices that best represent each parameter for each machine learning algorithm. Results: Only three echocardiographic parameters had no HRV index significantly correlated to them. The most frequently selected HRV index in the attribute selection process was the fractal short-term scaling exponent. The regression models (numeric parameters) reached reasonable performance (R > 0.5) for all except two parameters, while the classification models (categorical variables) achieved better performance, with precision and recall values higher than 0.74. Conclusion: HRV indices, both isolated and combined, are associated with cardiac morpho-functional properties in patients with CD, and may be used to predict echocardiographic parameters. Significance: The possibility of modeling the cardiac morpho-functional parameters in patients with CD using HRV indices opens the possibility to use HRV for risk assessment in patients with CD, especially those harboring the indeterminate form of the disease.