A new study in European Heart Journal – Digital Health showed that Us2.ai’s AI echo algorithm can interpret echocardiographic strain images with similar accuracy as conventional measurements, highlighting how AI can democratize cardiac strain exams and improve heart failure assessments.
- Echo strain imaging reflects myocardial deformation, and supports early heart failure diagnosis.
- However, echo strain workflows have plenty of room for improvement, given that current software uses semi-automated algorithms that “still require considerable human inputs” and measurements can vary widely with different operators.
An international research team used Us2.ai’s deep learning algorithm to measure LV global longitudinal strain (GLS) in echo exams from a real-world Taiwanese cohort (n=4,228 with & without HF) and a core-lab dataset from the PROMIS-HFpEF study (n=183 with HFpEF), and measured wall-motion abnormalities in a dataset from the HMC-QU-MI study (n= 158 with suspected MI). The Us2.ai echo AI workflow…
- Successfully analyzed 89%, 96%, and 98% of the datasets, omitting more exams from the Taiwanese cohort due to poor image quality
- Showed good agreement with manual measurements made with commercially available software (avg: −18.9 vs. −18.2 in Taiwanese cohort; −15.4 vs. −15.9 in PROMIS-HFpEF)
- Accurately identified patients with heart failure in the Taiwanese cohort (AUCs: 0.89 for total HF, 0.98 for HFrEF)
- Identified regional wall-motion abnormalities in the HMC-QU-MI study (AUC: 0.80)
These results were strong enough for the authors to suggest that AI echo solutions like Us2.ai’s could “democratize the use of cardiac strain measurements” in settings where resources or expertise are limited, improve echo lab efficiency, and even support challenging HFpEF and AMI diagnoses.
These results also highlight the clinical potential of AI-based GLS measurements, and perhaps helping clinicians overcome limitations with LVEF-based HF classification (e.g. with HFpEF)
- That last part is notable from an echo AI adoption perspective, given that CMS recently approved an outpatient reimbursement code for AI-based HFpEF detection ($285 per use)
- It could also prove to be notable for Us2.ai, which uniquely measures both LVEF and strain (among other measurements)
The Takeaway
Echo-based heart failure assessments are widely considered one of the strongest candidates for AI-driven clinical and efficiency improvements, and these results show how a solution that supports both LVEF and strain (along with the new CMS reimbursements for HFpEF) could help drive the echo AI adoption that everyone has been predicting.
