What is the clinical evidence from this study?

Study design: Other. Population: Cardiovascular disease (Myocardial Infarction) (n=15895). Intervention: xGNN4MI (Graph Neural Network). Primary outcome: Diagnostic superclass classification (AUC).

What does this research mean for the field?

xGNN4MI achieved an overall accuracy of 78% in classifying myocardial infarction subtypes on the PTB-XL dataset, compared to 62% on external validation data from the SHIP dataset. Novelty: ClaimNovelty.NOVEL_FINDING. Consensus alignment: ConsensusAlignment.CHALLENGES_CONSENSUS.

What question did this study set out to answer?

To improve the interpretability and explainability of AI models applied to cardiovascular disease prediction using ECG data.

February 8, 2026npj Digital Medicine0 citationsOpen Access

xGNN4MI: explainability of graph neural networks in 12-lead electrocardiography for cardiovascular disease classification

MMM. MäurerJustus-Liebig-Universität Gießen PHP. HempelUniversitätsmedizin Göttingen KSKristin SteinhausUniversitätsmedizin Göttingen

Key Result

The xGNN4MI framework achieved an AUC of 0.86 for diagnostic superclass classification and 0.92 for myocardial infarction localization, while identifying clinically relevant ECG leads.

Key Points

To improve the interpretability and explainability of AI models applied to cardiovascular disease prediction using ECG data.
Developed an open-source framework xGNN4MI for modeling ECG with graph neural networks.
Integrated explainable AI techniques for evaluating and visualizing the importance of ECG leads.
Tested framework on tasks including diagnostic superclass classification and localization of myocardial infarction.
Interpretable ECG-GNN models performed well on classification and localization tasks.
XAI analysis identified critical ECG leads impacting model decisions, providing insights into clinical relevance.
Demonstrated meaningful differentiation between types of myocardial infarction.

Structured PICO

Does the xGNN4MI framework provide interpretable and accurate cardiovascular disease classification and myocardial infarction localization from 12-lead ECGs?

Population

12-lead ECG datasets (PTB-XL and SHIP) for cardiovascular disease classification and myocardial infarction localization

Intervention

xGNN4MI (an open-source framework for graph neural networks in ECG modeling with explainable AI)

Outcome

Diagnostic superclass classification and localization of myocardial infarction

The xGNN4MI framework provides an interpretable graph neural network approach for 12-lead ECG classification, enabling systematic comparison of AI decision-making with established clinical knowledge.

Limitations

GNNExplainer assumes influential subgraphs are structurally connected, potentially missing distant relationships.
Does not explicitly model temporal dynamics.
Single-label formulation does not fully capture real-world comorbidities.

Abstract

Abstract The clinical deployment of artificial intelligence (AI) solutions for assessing cardiovascular disease (CVD) risk in 12-lead electrocardiography (ECG) is hindered by limitations in interpretability and explainability. To address this, we present xGNN4MI, an open-source framework for graph neural networks (GNNs) in ECG modeling for interpretable CVD prediction. Our framework facilitates modeling clinically relevant spatial relationships between ECG leads and their temporal dynamics. We integrated explainable AI (XAI) and developed a task-specific XAI evaluation and visualization workflow to identify ECG leads crucial to the model’s decision-making process, enabling a systematic comparison with established clinical knowledge. We evaluated xGNN4MI on two challenging tasks: diagnostic superclass classification and localization of myocardial infarction. Our findings show that the interpretable ECG-GNN models demonstrate good performance across the tasks. XAI analysis revealed clinically meaningful training effects, such as differentiating between anteroseptal and inferior myocardial infarction. Our work demonstrates the potential of ECG-GNNs for providing trustworthy and interpretable AI-based CVD diagnosis.

Ask AI

Helpful

Bookmark

View Full Paper