Article | Nov 22, 2023

The first fully automated AI software with validated global longitudinal strain in patients with and without atrial fibrillation; plus regional strain validated in real-world datasets; plus accurate identification of patients with heart failure, as well as those with regional wall-motion abnormalities, published in the European Heart Journal's Digital Health.

Key Findings

  • This is the first external validation of a fully automated AI GLS algorithm across patients with and without atrial fibrillation, addressing a major gap since AF degrades strain measurement quality and was excluded from most prior AI strain studies.
  • In the Taiwanese cohort, automated GLS accurately identified patients with heart failure with an AUC of 0.89 for total HF and AUC of 0.98 for HFrEF, demonstrating strong clinical utility for AI-derived strain in HF detection.
  • In the HMC-QU-MI cohort, automated regional strain identified regional wall-motion abnormalities with an average AUC of 0.80, validating its use for ischaemia detection beyond global function.
  • Deep learning algorithms interpreted echocardiographic strain images with similar accuracy to conventional vendor-specific measurements across multiple independent external cohorts and imaging systems, supporting vendor-neutral clinical deployment.
  • The results highlight the potential for AI strain measurements to reduce time, cost, and variability in echo labs globally, particularly by removing the need for vendor-specific software and manual contouring.

 

AI Automated Strain

Myhre, P. L., Hung, C., Frost, M., Jiang, Z., Ouwerkerk, W., Teramoto, K., Svedlund, S., Saraste, A., Hage, C., Tan, R., Beussink-Nelson, L., Fermer, M. L., Gan, L., Hummel, Y. M., Lund, L. H., Shah, S. J., Lam, C. S.P. & Tromp, J. (2023). External validation of a deep learning algorithm for automated echocardiographic strain measurements. European Heart Journal. https://doi.org/10.1093/ehjdh/ztad072