Clinical value of automated left atrial strain analysis based on deep learning algorithms in echocardiography for early diagnosis of heart failure with preserved ejection fraction

European Journal of Medical Research

卷:31

Research

Libo Wang¹  Haiyang Mu²  Ye Tian³  Ying Lin⁴  Jian Wang⁴  Liye Zhang¹  Xingying Wu¹  Mei Chen¹  Meiling Yang¹  Chaofei Zhou⁴ 

[1] Department of Cardiology, Hainan Hospital of Chinese PLA General Hospital, 572013, Sanya, Hainan, China

[2] Department of Medical Engineering, Hainan Hospital of Chinese PLA General Hospital, 572013, Sanya, Hainan, China

[3] Department of Ultrasound, Hainan Hospital of Chinese PLA General Hospital, 572013, Sanya, Hainan, China

[4] Department of Emergency Medicine, Hainan Hospital of Chinese PLA General Hospital, 572013, Sanya, Hainan, China

关键词: Heart failure with preserved ejection fraction;    Left atrial strain;    Deep learning;    Echocardiography;    Diastolic dysfunction;    Speckle-tracking echocardiography;   

DOI  :  10.1186/s40001-026-04802-1

 received in 2026-03-02, accepted in 2026-06-18,  发布年份 2026

来源: Springer

【 摘 要 】

BackgroundHeart failure with preserved ejection fraction (HFpEF) remains challenging to diagnose due to the absence of overt systolic dysfunction. Left atrial reservoir strain (LARS) has emerged as a sensitive marker of left atrial dysfunction, yet its clinical application is constrained by measurement variability and time demands. We investigated the diagnostic performance of automated deep learning (DL)-derived LARS for early HFpEF identification.MethodsThis retrospective observational study enrolled 200 patients (107 HFpEF, 93 controls) who underwent transthoracic echocardiography. HFpEF was diagnosed per 2021 ESC guidelines. All echocardiographic datasets were analyzed using a validated DL-based automated platform (EchoGo Pro, Ultromics). Agreement between DL-derived and manually measured parameters was evaluated by Pearson correlation and Bland–Altman analysis. Diagnostic performance was assessed by receiver operating characteristic (ROC) curve analysis, and independent predictors of HFpEF were identified via multivariate logistic regression.ResultsDL-derived LARS was significantly lower in HFpEF patients compared to controls (19.3 ± 5.7% vs. 30.8 ± 6.4%, P < 0.001). Strong agreement was demonstrated between DL-derived and manually measured LARS (r = 0.911; mean bias − 0.67 ± 2.34%), with superior reproducibility of the automated approach (ICC = 0.974). DL-derived LARS achieved an AUC of 0.806 (95% CI 0.748–0.864), sensitivity of 81.3%, and specificity of 83.9% at a cutoff of 24.2%, comparable to manually derived LARS (AUC = 0.818, P = 0.631) and numerically higher than LAVI and E/e' ratio, although these pairwise differences did not reach statistical significance. On multivariate analysis, DL-derived LARS was the strongest independent echocardiographic predictor of HFpEF (adjusted OR = 0.74, 95% CI 0.65–0.84, P < 0.001).ConclusionsAutomated DL-based LARS analysis provides accurate, reproducible, and time-efficient quantification comparable to expert manual measurement, demonstrating robust diagnostic value for early HFpEF identification.

【 授权许可】

CC BY-NC-ND   
© The Author(s) 2026