Evaluation of a Low-Cost Single-Lead ECG Module for Vascular Ageing Prediction and Studying Smoking-induced Changes in ECG

TL;DR

This study demonstrates that a low-cost single-lead ECG module can accurately predict vascular age and detect smoking-related ECG changes, offering a sustainable and accessible healthcare tool for young adults.

Contribution

The paper introduces a novel low-cost single-lead ECG device combined with machine learning models for vascular age prediction and smoking impact analysis, with transfer learning enhancing accuracy.

Findings

Random forest achieved MSE of 0.07 and R2 of 0.99 for vascular age prediction.

Smoking significantly affects ECG features identified by explainable AI.

Transfer learning improved model performance on ECG data.

Abstract

Vascular age is traditionally measured using invasive methods or through 12-lead electrocardiogram (ECG). This paper utilizes a low-cost single-lead (lead-I) ECG module to predict the vascular age of an apparently healthy young person. In addition, we also study the impact of smoking on ECG traces of the light-but-habitual smokers. We begin by collecting (lead-I) ECG data from 42 apparently healthy subjects (smokers and non-smokers) aged 18 to 30 years, using our custom-built low-cost single-lead ECG module, and anthropometric data, e.g., body mass index, smoking status, blood pressure, etc. Under our proposed method, we first pre-process our dataset by denoising the ECG traces, followed by baseline drift removal, followed by z-score normalization. Next, we create another dataset by dividing the ECG traces into overlapping segments of five-second duration. We then feed both segmented and unsegmented datasets to a number of machine learning models, a 1D convolutional neural network, and ResNet18 model, for vascular ageing prediction. We also do transfer learning whereby we pre-train our models on a public PPG dataset, and later, fine-tune and evaluate them on our unsegmented ECG dataset. The random forest model outperforms all other models and previous works by achieving a mean squared error (MSE) of 0.07 and coefficient of determination R2 of 0.99, MSE of 3.56 and R2 of 0.26, MSE of 0.99 and R2 of 0.87, for segmented ECG dataset, for unsegmented ECG dataset, and for transfer learning scenario, respectively. Finally, we utilize the explainable AI framework to identify those ECG features that get affected due to smoking. This work is aligned with the sustainable development goals 3 and 10 of the United Nations which aim to provide low-cost but quality healthcare solutions to the unprivileged. This work also finds its applications in the broad domain of forensic science.