Research hotspots and development trends of machine learning in the field of atrial fibrillation

CHEN Rui; ZHANG Ting; HUANG Han; CAO Wenzhai

doi:10.3969/j.issn.1002-266X.2026.01.017

Abstract：Objective To analyze research hotspots and development trends in the application of machine learning(ML)in the field of atrial fibrillation(AF).Methods Literature related to ML in AF from January 1,2014,to Decem-ber 31,2024,was retrieved from the Web of Science database.A total of 880 publications were finally included,compris-ing 779 articles and 101 reviews.CiteSpace software(version 6.4.R1)was used to perform cluster analysis on these 880 publications,and we examined the characteristics including publication volume,contributing countries/institutions/au-thors,cited authors/journals/articles,as well as keyword co-occurrence,clustering,and burst detection.Results The r² of the publication growth regression curve was 0.9563,indicating an exponential increase in ML-related AF literature.Publications were relatively scarce from 2014 to 2017,followed by explosive growth after 2018,with 2024 having the high-est output.China,the USA,and the UK were the top three contributing countries.The top five contributing institutions were University of London,Harvard University,Mayo Clinic,University of Liverpool,and Liverpool Heart and Chest Hos-pital.Gregory Y.H.Lip(University of Liverpool)was the most prolific author,while Gerhard Hindricks was the most fre-quently cited author.Frontiers in Cardiovascular Medicine published the most articles,Circulation was the most cited jour-nal,and Cardiovascular Research had the highest centrality.Among the top 10 cited references,three were from the USA,two from Germany,and one each from Denmark,China,the UK,Canada,and Greece.Excluding two AF guidelines,the remaining eight research articles all utilized deep learning models.The top five highest-frequency keywords were atrial fi-brillation,machine learning,risk,artificial intelligence,and catheter ablation.Keyword clustering yielded seven clus-ters,primarily focusing on four aspects:machine learning methodologies,data sources,risk assessment,and clinical deci-sion-making.The keyword with the strongest and longest burst intensity was"arrhythmia",while bursts for"risk predic-tion"and"expression"persisted from their onset to the present.Conclusions Research interest in applying ML to AF is increasing annually.Current hotspots concentrate on deep learning models,electrocardiogram and imaging feature analy-sis,and disease prediction/identification.Future research is likely to focus on key areas such as genetic prediction of AF,patient risk prediction,and prognosis following catheter ablation.