External validation of a deep learning algorithm for automated echocardiographic strain measurements.

Area of Science:

• Cardiology
• Medical Imaging
• Artificial Intelligence

Background:

• Echocardiographic strain imaging assesses myocardial deformation, offering sensitive insights into cardiac function and wall-motion abnormalities.
• Deep learning (DL) presents an opportunity to automate the complex interpretation of echocardiographic strain imaging.

Purpose of the Study:

• To develop and validate a DL-based algorithm for automated left ventricular (LV) strain measurements.
• To assess the accuracy of automated GLS and regional strain in identifying heart failure (HF) and regional wall-motion abnormalities.

Main Methods:

• An automated DL algorithm was developed and trained on an internal dataset.
• Global longitudinal strain (GLS) was externally validated in a Taiwanese cohort and the PROMIS-HFpEF study.
• Regional strain was validated in the HMC-QU-MI study for patients with suspected myocardial infarction.

Main Results:

• The DL workflow successfully analyzed a high percentage of studies across all cohorts (89-98%).
• Automated GLS demonstrated good agreement with manual measurements in both the Taiwanese cohort (R=0.84) and PROMIS-HFpEF (R=0.76).
• Automated GLS accurately identified HF (AUC=0.89) and regional strain identified wall-motion abnormalities (AUC=0.80).

Conclusions:

• DL algorithms can interpret echocardiographic strain images with accuracy comparable to conventional methods.
• The study highlights the potential of DL to democratize cardiac strain measurements, reducing global costs and time for echocardiography laboratories.