Jacc Advances: Automated Echocardiographic Detection of HFpEF using AI

Ashley P. Akerman PhD, Mihaela Porumb PhD, Christopher G. Scott MS, Arian Beqiri PhD, Agisilaos Chartsias PhD, Alexander J. Ryu MD, William Hawkes PhD, Geoffrey D. Huntley MD, Ayana Z. Arystan MD, Garvan C. Kane MD, Sorin V. Pislaru MD, Francisco Lopez-Jimenez MD, Alberto Gomez PhD, Rizwan Sarwar MRCP, Jamie O’Driscoll PhD, Paul Leeson MB, Ross Upton PhD, Gary Woodward PhD, Patricia A. Pellikka MD.

Background

Detection of heart failure with preserved ejection fraction (HFpEF) involves integration of multiple imaging and clinical features which are often discordant 
or indeterminate. 

Objectives

We applied artificial intelligence (AI) to analyze a single apical fourchamber (A4C) transthoracic echocardiogram videoclip to detect HFpEF.

Methods

A three-dimensional convolutional neural network was developed and trained on A4C videoclips to classify patients with HFpEF (diagnosis of HF, EF≥50%, and echocardiographic evidence of increased filling pressure; cases) versus without 
HFpEF (EF≥50%, no diagnosis of HF, normal filling pressure; controls). Model outputs were classified as HFpEF, no HFpEF, or non-diagnostic (high uncertainty). Performance was assessed in an independent multi-site dataset and compared to previously validated clinical scores.

Results

Training and validation included 2971 cases and 3785 controls (validation holdout, 16.8% patients), and demonstrated excellent discrimination (AUROC:0.97 
[95%CI:0.96-0.97] and 0.95 [0.93-0.96] in training and validation, respectively).

In independent testing (646 cases, 638 controls), 94 (7.3%) were non-diagnostic; sensitivity (87.8%; 84.5-90.9%) and specificity (81.9%; 78.2-85.6%) were maintained in clinically relevant subgroups, with high repeatability and reproducibility.

EchoGo® Heart Failure demonstrated high sensitivity and specificity, detecting 87.8% of patients who had HFpEF, and 81.9% of patients that did not. These results exceed what is usually observed in routine clinical practice. 

Of 701 and 776 indeterminate outputs from the HFA-PEFF and H2FPEF scores, the AI HFpEF model correctly reclassified 73.5 and 73.6%, respectively. 

EchoGo® Heart Failure had fewer non-diagnostic outputs compared to HFA-PEFF and correctly reclassified 515 (73.5%) of patients. 

EchoGo® Heart Failure had fewer non-diagnostic outputs compared to H2FPEF and  correctly reclassified 571 (73.6%) of patients.  

During follow-up (median [IQR]:2.3 [0.5-5.6] years), 444 (34.6%) patients died; mortality was higher in patients classified as HFpEF by AI (hazard ratio [95%CI]:1.9 [1.5-2.4])

Age-adjusted risk of mortality was higher when patients received a positive diagnostic output (HFpEF) from the AI HFpEF model compared to negative diagnostic output (not HFpEF).

Conclusion

An AI HFpEF model based on a single, routinely acquired
echocardiographic video demonstrated excellent discrimination of patients with versus without HFpEF, more often than clinical scores, and identified patients with higher mortality.