Leveraging Visibility Graphs for Enhanced Arrhythmia Classification with Graph Convolutional Networks

TL;DR

This work investigates transforming ECG beats into graphs via Visibility Graphs (VG) and Vector Visibility Graphs (VVG) and classifying arrhythmias with Graph Convolutional Networks (GCNs) under the ANSI/AAMI EC57 standard using MIT-BIH data. The study systematically compares multiple GCN architectures and segmentation/feature strategies, showing that VG and VVG mappings enable direct, preprocessing-free classification from raw signals, with VG offering better efficiency and VVG benefiting from multi-lead information. Across inter- and intra-patient paradigms, simpler GCNs often outperform heavier variants, and the multi-lead VVG approach yields notable gains at the expense of higher computational load. While classifying supraventricular ectopic beats (S) remains challenging in inter-patient splits, intra-patient experiments show strong performance, and the method demonstrates competitive results against baselines, highlighting the potential of graph-based ECG analysis with minimal preprocessing.

Abstract

Arrhythmias, detectable through electrocardiograms (ECGs), pose significant health risks, underscoring the need for accurate and efficient automated detection techniques. While recent advancements in graph-based methods have demonstrated potential to enhance arrhythmia classification, the challenge lies in effectively representing ECG signals as graphs. This study investigates the use of Visibility Graph (VG) and Vector Visibility Graph (VVG) representations combined with Graph Convolutional Networks (GCNs) for arrhythmia classification under the ANSI/AAMI standard, ensuring reproducibility and fair comparison with other techniques. Through extensive experiments on the MIT-BIH dataset, we evaluate various GCN architectures and preprocessing parameters. Our findings demonstrate that VG and VVG mappings enable GCNs to classify arrhythmias directly from raw ECG signals, without the need for preprocessing or noise removal. Notably, VG offers superior computational efficiency, while VVG delivers enhanced classification performance by leveraging additional lead features. The proposed approach outperforms baseline methods in several metrics, although challenges persist in classifying the supraventricular ectopic beat (S) class, particularly under the inter-patient paradigm.

Figures (6)

• Figure 1: Example of the VG method application.
• Figure 2: Example of the VVG method application. (a) Multivariate time series represented as two ECG signals from different leads. (b) Graph generated by the VVG. (c) Visibility among the vectors. Source: Adapted from ren2019vector.
• Figure 3: Flowchart of our proposed methodology.
• Figure 4: Example of mapping heartbeats from one lead (patient 118) corresponding to classes N, S, and V using the VG method.
• Figure 5: Proposed CNN architectures.