Using explainable AI to investigate electrocardiogram changes during healthy aging -- from expert features to raw signals

TL;DR

This study investigates how healthy aging alters ECG signals by comparing a feature-based XGBoost approach with a deep-learning XResNet50 model trained on raw ECGs. Using the Autonomic Aging dataset, the authors apply SHAP and beat-aligned saliency maps to reveal explainable, age-discriminative patterns, finding a decline in inferred breathing rate and rising SDANN5 with age in the feature-based analysis, and a central role for the P-wave in age prediction for raw signals. The two methods achieve comparable predictive performance, highlighting complementary information captured by long-range versus short-range ECG cues. These insights advance understanding of aging-related cardiac changes and demonstrate the value of explainable AI for interpreting ECG-based age predictions, with potential implications for early cardiovascular risk detection.

Abstract

Cardiovascular diseases remain the leading global cause of mortality. Age is an important covariate whose effect is most easily investigated in a healthy cohort to properly distinguish the former from disease-related changes. Traditionally, most of such insights have been drawn from the analysis of electrocardiogram (ECG) feature changes in individuals as they age. However, these features, while informative, may potentially obscure underlying data relationships. In this paper we present the following contributions: (1) We employ a deep-learning model and a tree-based model to analyze ECG data from a robust dataset of healthy individuals across varying ages in both raw signals and ECG feature format. (2) We use explainable AI methods to identify the most discriminative ECG features across age groups.(3) Our analysis with tree-based classifiers reveals age-related declines in inferred breathing rates and identifies notably high SDANN values as indicative of elderly individuals, distinguishing them from younger adults. (4) Furthermore, the deep-learning model underscores the pivotal role of the P-wave in age predictions across all age groups, suggesting potential changes in the distribution of different P-wave types with age. These findings shed new light on age-related ECG changes, offering insights that transcend traditional feature-based approaches.

Figures (5)

• Figure 1: Age-group distribution. Age-group distribution in terms of age groups provided in the Autonomic Aging datasetdataset_paper. The age groups span a range from 18 to 92 years, where the majority of patients are between 20 to 50 years old.
• Figure 2: Predictive performance. Predictive performance results of the models per age group in terms of AUC on the test set, where the yellow (left) age group represents the XGBoost and the right (blue) the XResNet.
• Figure 3: SHAP values across age groups. The 10 most important features for classifying the age groups are depicted in these subplots. In each subplot, the features are arranged in descending order of importance, emphasizing their significance in age group classification. The color scheme, with blue dots representing low feature values and red dots denoting high feature values, provides a visual representation of the feature’s influence across different age groups. As you move from left to right and from top to bottom, you explore the SHAP values for all age groups.
• Figure 4: Aggregated mean heartbeat. Aggregated mean heartbeat for all age groups showcases ECG feature trends across age groups.
• Figure 5: Saliency maps across age groups. Within each subplot, as you progress from left to right and from top to bottom, you navigate through the beat-level ECG saliency maps, shedding light on the key ECG features that contribute to age group differentiation according to the color map scheme. ’Subjects’ and ’Heartbeats’ state the number of subjects and heartbeats used to create these plots. The eight highest gradients are marked in red.