AI-assisted focused cardiac ultrasound (FoCUS) is on par with transthoracic echocardiogram (TTE) when it comes to assessing left ventricular ejection fraction (LVEF) estimations, according to research published October 28 in npj Digital Medicine.
A team led by Pouya Motazedian, MD, from the University of Ottawa Heart Institute in Ottawa, Ontario, Canada found that AI-FoCUS had high area under the curve (AUC) values when identifying an abnormal LVEF and severe dysfunction.
“We report that FoCUS AI-assisted LVEF assessments provide highly reproducible LVEF estimations in comparison to formal TTE,” the Motazedian team wrote.
While TTE is the standard for determining LVEF in clinical practice, the researchers noted that it is “often not readily available” for immediate bedside evaluation and in low-resource settings. Assessing LVEF is fundamental in FoCUS exams.
While previous studies suggest that FoCUS is useful in screening for LVEF, the researchers pointed out that these have been limited to trained sonographers and clinicians with formal training. AI has also demonstrated promise in this area, but the researchers again noted that studies with AI have involved formally trained, experienced clinicians and sonographers.
Motazedian and colleagues sought to test the accuracy of FoCUS AI-assisted LVEF assessment and compare this between novice and experienced users. They used a commercially available AI device (Kosmos, EchoNous) to perform FoCUS exams and included data from 449 participants with 424 exams.
The team found that the overall intraclass coefficient was 0.904 (with 1 as reference). This included a coefficient of 0.845 in the experienced cohort and 0.921 in the novice cohort.
The team also reported a significant bias of 0.73% toward TTE (p = 0.005), with a level of agreement of 11.2%. Additionally, it found that the categorical grading of LVEF severity had excellent agreement with TTE (weighted kappa = 0.83).
The researchers also found that AI-FoCUS had high AUC values for identifying an abnormal LVEF and severe dysfunction.
| AI-FoCUS performance for LVEF assessment | ||
|---|---|---|
| Abnormal LVEF (< 50%) | Severe dysfunction (< 30%) | |
| AUC | 0.98 | 0.99 |
| Sensitivity | 92.8% | 78.1% |
| Specificity | 92.3% | 98% |
| Positive predictive value | 0.83 | 0.76 |
| Negative predictive value | 0.97 | 0.98 |
The study authors suggested that based on their results, AI-assisted FoCUS evaluation could serve as a surrogate for a formal TTE in this area in both inpatient and outpatient settings.
“The integration of AI into FoCUS remains a potential method for minimally trained users to accurately and efficiently identify abnormal LVEF, while simultaneously providing a reliable evaluation of severity,” they wrote.
The authors added that given the potential changes to management depending on the degree of severity, “AI-FoCUS may provide a rapid, accurate assessment of LVEF and therefore allow for prompt diagnosis and management.”
The full study can be found here.
