Bellfield, RAA, Rendon Hormiga, P, Olier, I, Lotto, RR, Jones, ID, Lip, GYH and Ortega-Martorell, S ORCID: 0000-0001-9927-3209
AI-driven clustering and visualisation of ECG signals to enhance screening for atrial fibrillation: The supermarket/hypermarket opportunistic screening for atrial fibrillation (SHOPS-AF) study.
Heart Rhythm O2.
(Accepted)
![]() |
Text
Enhancing opportunistic AF screening with AI.pdf - Accepted Version Access Restricted Available under License Creative Commons Attribution Non-commercial No Derivatives. Download (1MB) |
![]() |
Text
ShopsAF and GTM Supplementary Materials.docx - Supplemental Material Access Restricted Available under License Creative Commons Attribution Non-commercial No Derivatives. Download (482kB) |
Abstract
Background: Atrial fibrillation (AF) is commonest arrhythmia globally, associated with an increased risk of serious health issues. As its prevalence rises, healthcare systems face significant challenges, including escalating treatment costs and the inherent difficulties of detecting AF, particularly in paroxysmal cases where symptoms are intermittent. Objective: This study investigates the application of unsupervised machine learning, specifically generative topographic mapping (GTM), to support AF screening and risk stratification. Methods: The SHOPS-AF study deployed single-lead ECG sensors (MyDiagnostick) embedded in supermarket trolley handles across four sites in Northwest England. This community-based approach successfully engaged the public in opportunistic AF screening. However, diagnosis was limited by reliance on transient ECG recordings. To improve analysis, we selected a subset of 97 ECG traces (78 for training and 19 for testing) reviewed by a consultant cardiologist, comprising AF (n=23), possible AF (n=9), and normal rhythm (n=65). From these, 477 20-second ECG snippets were extracted to train the GTM model. Results: The GTM generated interpretable membership maps, clustering ECG snippets into visually distinct regions with similar features. These maps enable clinicians to explore heart rhythm dynamics over time and track patient trajectories across risk states. Conclusions: This study demonstrates the potential of our proposed methodology to uncover latent patterns in ECG data, providing deeper insights into individual heart rhythm patterns, and supporting more nuanced AF risk assessment and the overall effectiveness of AF detection and management. By embedding interpretable AI into screening tools, we aim to improve early detection and reduce the clinical burden of AF.
Item Type: | Article |
---|---|
Subjects: | Q Science > QA Mathematics > QA75 Electronic computers. Computer science R Medicine > R Medicine (General) |
Divisions: | Computer Science and Mathematics Nursing and Advanced Practice |
Publisher: | Elsevier |
Date of acceptance: | 7 July 2025 |
Date Deposited: | 10 Jul 2025 10:25 |
Last Modified: | 10 Jul 2025 10:30 |
URI: | https://researchonline.ljmu.ac.uk/id/eprint/26753 |
![]() |
View Item |