# A Screening Method for Determining Left Ventricular Systolic Function Based on Spectral Analysis of a Single-Channel Electrocardiogram Using Machine Learning Algorithms

## TL;DR

This paper introduces a machine learning-based method to screen for heart dysfunction using a single ECG channel, achieving high diagnostic accuracy.

## Contribution

A novel screening method for left ventricular systolic dysfunction using single-channel ECG and machine learning is proposed.

## Key findings

- Lasso regression achieved 79.2% sensitivity and 81.7% specificity for detecting reduced ejection fraction.
- Extra Trees model showed 83.1% sensitivity and 82.7% specificity for severe ejection fraction reduction.
- External testing confirmed 98% accuracy, 98.4% specificity, and 93.5% sensitivity.

## Abstract

Background and Objectives: Given the non-specificity of symptoms and complex methods for diagnosing heart failure, which are not applicable in screening, it is of great importance to develop a simple screening method for identifying systolic dysfunction of the heart based on available biosignals, one of which is a single-channel electrocardiogram (ECG). The method does not require the participation of medical staff. Aim: To create a screening model for detecting left ventricular systolic dysfunction in a complex analysis of single-channel ECG parameters using machine learning algorithms Methods: We included 624 patients aged 18 to 90 years. All patients underwent echocardiography and single-channel I-lead ECG recording using a portable electrocardiograph. The left ventricle ejection fraction (LV EF) was determined in the apical 2-chamber and 4-chamber view using the BIPLANE Simpson method and confirmed by two independent experts. Single-channel ECG analysis was performed using advanced signal processing and machine learning techniques. Results: For identifying LV EF below 52% in men and below 54% in women, the best result was demonstrated by “Lasso regression”: sensitivity 79.2%, specificity 81.7%, AUC = 0.849. For detection of LVEF below 40%, the “Extra Trees” model was the best, with a sensitivity of 83.1% and a specificity of 82.7%, AUC = 0.972. External testing of the algorithm was conducted on a sample of 600 patients. The accuracy was 98%, specificity 98.4%, and sensitivity 93.5%. Conclusions: The results indicate quite high diagnostic accuracy of screening for left ventricular systolic dysfunction when analyzing single-channel ECG parameters using modern signal processing and machine learning technologies.