ORIGINAL RESEARCH article
Front. Med.
Sec. Pathology
Volume 12 - 2025 | doi: 10.3389/fmed.2025.1661696
This article is part of the Research TopicFrom Black-box to Clarity in Lesion Diagnostics: Clinical Causal Cognition Led Interpretable Diagnostic AI SystemsView all articles
A New Integrated Machine Learning Model: Application to Improve the Accuracy of Predicting Left Atrial Appendage Thrombus in Patients with Non-Valvular Atrial Fibrillation
Provisionally accepted
- 1Xinxiang Medical University, Xinxiang, China
- 2Luoyang Central Hospital Affiliated to Zhengzhou University, Luoyang, China
- 3Second Affiliated Hospital of Xi'an Jiaotong University, Xi'an, China
- 4Yizhun Medical AI Co Ltd, Beijing, China
Select one of your emails
You have multiple emails registered with Frontiers:
Notify me on publication
Please enter your email address:
If you already have an account, please login
You don't have a Frontiers account ? You can register here
Background: Non-Valvular Atrial fibrillation (NVAF) and atrial flutter are significant contributors to left atrial appendage thrombus (LAAT) formation. This study explores the potential of machine learning (ML) models integrating transthoracic echocardiography (TTE) and clinical data for non-invasive LAAT detection and risk assessment. Methods: A total of 698 patients with NVAF was recruited from Luoyang Central Hospital between January 2021 and May 2024, including 558 patients for retrospective analysis and 140 for prospective validation. Based on transesophageal echocardiography (TEE) results, patients were categorized into three groups: non-thrombotic, blood stasis, and thrombotic. Four ML algorithms-Random Forest, Logistic Regression (LR), Support Vector Machine, and XGBoost-were developed using TTE and clinical data to predict LAAT.Results: Univariate analysis identified significant predictors of LAAT, including permanent AF, heart failure, BNP, uric acid, D-dimer, mitral regurgitation, LVEF, LVED, LAD, CHA₂DS₂-VASc score, and LAA velocity (p < 0.05). The combined TTE data model outperformed independent TTE-based models but was slightly less accurate than the TEE model. Among ML algorithms, the LR model demonstrated the best performance, achieving an area under the curve (AUC) of 80.9% in the test set and 78.7% in prospective validation for the thrombotic state group. For the thrombotic group, the LR model achieved an AUC of 80.0%, closely approaching the TEE model's 84.0%. Conclusion: The LR model provides a reliable non-invasive approach for LAAT screening in high-risk AF patients by integrating TTE features with clinical data, potentially reducing reliance on TEE.
Keywords: non-valvular atrial fibrillation (NVAF), Transthoracic echocardiography (TTE), Left Atrial Appendage Thrombosis (LAAT), machine learning (ML) models, Transesophageal echocardiography (TEE)
Received: 08 Jul 2025; Accepted: 05 Aug 2025.
Copyright: © 2025 Wang, Mai, Huo, Li, Zhou, He and Li. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.
* Correspondence: Hua Wang, Xinxiang Medical University, Xinxiang, China
Disclaimer: All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article or claim that may be made by its manufacturer is not guaranteed or endorsed by the publisher.