Automated Echocardiographic Detection of Heart Failure With Preserved Ejection Fraction Using Artificial Intelligence

Akerman, AP; Porumb, M; Scott, CG; Beqiri, A; Chartsias, A; Ryu, AJ; Hawkes, W; Huntley, GD; Arystan, AZ; Kane, GC; Pislaru, SV; Lopez-Jimenez, F; Gomez, A; Sarwar, R; O'Driscoll, J; Leeson, P; Upton, R; Woodward, G; Pellikka, PA

doi:10.1016/j.jacadv.2023.100452

Heart failure & cardiomyopathies

Automated Echocardiographic Detection of Heart Failure With Preserved Ejection Fraction Using Artificial Intelligence

JACC AdvPublished 28 Jul 2023Diagnostic accuracy study

Read the original article ↗PubMed

Cite

Authors' conclusion

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

Study context

Population: Patients with EF ≥50% with heart failure with preserved ejection fraction (HFpEF) (HF diagnosis, increased filling pressure; cases) vs without HFpEF (no HF diagnosis, normal filling pressure; controls)test set 646 cases, 638 controls

Comparison: Artificial intelligence (AI) HFpEF model vs clinical scores
Outcome: Detection of HFpEF (discrimination, sensitivity, specificity) in independent testing

Methodological detailsindependent multisite test data set

Independent multisite test data set

Performance

As reported in the abstract

Detection of HFpEF by the AI model, independent multisite test set

Sensitivity

87.8%

95% CI 84.5%–90.9%

Specificity

81.9%

95% CI 78.2%–85.6%

Nondiagnostic outputs, n (%)

94

7.3%

Further findings

All-cause mortality, AI-classified HFpEF vs not · median follow-up 2.3 (IQR 0.5-5.6) years

	← Lower risk ←→ Higher risk →	HR (95% CI)
AI-classified HFpEF	0.512	1.9 (1.5–2.4)
	0.512Hazard ratio (log scale)

All-cause mortality, AI-classified HFpEF vs not — Deaths, n (%): 444 (34.6%)

Discrimination (AUC) in training and validation

AUC, training: 0.97 (95% CI 0.96-0.97)
AUC, validation: 0.95 (95% CI 0.93-0.96)

2,971 cases and 3,785 controls; validation holdout 16.8% of patients

Reclassification of indeterminate HFA-PEFF and H2FPEF outputs

Correctly reclassified, HFA-PEFF indeterminate (n = 701): 73.5%
Correctly reclassified, H2FPEF indeterminate (n = 776): 73.6%

Reference

Authors: Akerman AP, Porumb M, Scott CG, Beqiri A, Chartsias A, Ryu AJ, et al.
All 19 authors
Akerman AP, Porumb M, Scott CG, Beqiri A, Chartsias A, Ryu AJ, Hawkes W, Huntley GD, Arystan AZ, Kane GC, Pislaru SV, Lopez-Jimenez F, Gomez A, Sarwar R, O'Driscoll J, Leeson P, Upton R, Woodward G, Pellikka PA
Journal: JACC: Advances
Published: 28 Jul 2023
DOI: 10.1016/j.jacadv.2023.100452
PubMed: 38939447

Topic: Heart failure & cardiomyopathies
Also indexed: Cardiovascular imaging

Comparison

AI HFpEF model (3D convolutional neural network on a single apical 4-chamber echo clip) vs clinical scores (HFA-PEFF, H2FPEF)

Diagnostic definitions

AI HFpEF model: 3-dimensional convolutional neural network on apical 4-chamber video clips; outputs HFpEF, no HFpEF or nondiagnostic
HFA-PEFF: previously validated clinical score
H2FPEF: previously validated clinical score

Key notes

Test set: 646 cases, 638 controls
Maintained in clinically relevant subgroups, with high repeatability and reproducibility

Original publication

Automated Echocardiographic Detection of Heart Failure With Preserved Ejection Fraction Using Artificial Intelligence. JACC Adv. Published 28 July 2023. doi:10.1016/j.jacadv.2023.100452

Read the original article ↗

Composed by AI from the published abstract and checked against it. How Briefs are made →