Quantifying Cardiovascular Wellbeing Through ECG Age: Acquisition Constraints and the Case for Handcrafted Features.

Area of Science:

• Cardiology
• Biomedical Engineering
• Artificial Intelligence

Background:

• Electrocardiogram (ECG) derived age, the difference between estimated and chronological age, is a marker for cardiovascular health.
• Increased ECG age delta correlates with cardiovascular conditions and vascular aging.
• Limited research exists on ECG acquisition parameter influence on neural network age estimation.

Purpose of the Study:

• To comprehensively evaluate the impact of ECG acquisition parameters on neural network performance for ECG age estimation.
• To compare deep neural networks (DNNs) with fully connected networks (FCNs) utilizing handcrafted features.
• To highlight the utility and significance of ECG feature engineering.

Main Methods:

• Conducted a comprehensive study on ECG acquisition parameters' influence on neural network performance.
• Compared DNNs against FCNs trained with state-of-the-art handcrafted ECG features.
• Evaluated performance metrics for both approaches.

Main Results:

• Handcrafted ECG features demonstrate competitive performance against end-to-end DNNs for age estimation.
• Handcrafted features require minimal computational resources and offer explainability.
• Identified key ECG parameters relevant to cardiovascular well-being through feature analysis.

Conclusions:

• ECG age is an intuitive metric for cardiovascular well-being, useful for patient and physician understanding.
• Optimizing ECG acquisition parameters can aid in developing better ECG-based tools for diverse platforms.
• Handcrafted features present a viable, explainable, and resource-efficient alternative for ECG age estimation.