AI interpretation of angiogram videos may help estimate LVEF
Click Here to Manage Email Alerts
Key takeaways:
- Artificial intelligence analysis of angiogram videos may help identify reduced left ventricular ejection fraction.
- The AI may provide a “noninvasive alternative to left ventriculography.”
A deep neural network demonstrated good discrimination of reduced left ventricular ejection fraction using angiogram videos compared with transthoracic echocardiogram, researchers reported.
“Researchers at UCSF developed CathEF, an AI model, which can estimate the LVEF from standard coronary angiograms,” Robert Avram, MD, MSc, cardiologist in the health division of cardiology at the University of California, San Francisco (UCSF), and an interventional cardiologist at the Montreal Heart Institute, University of Montreal, told Healio. “This measure is important for heart function assessment and patient management. The study results show that CathEF could accurately predict LVEF and its performance remained robust across various clinical conditions and patient demographics. This suggests a promising potential for AI in enhancing cardiac care and decision-making.”
Image: Adobe Stock
AI to identify reduced LVEF
Avram and colleagues trained and tested a video-based deep neural network called CathEF to discriminate reduced LVEF and predict LVEF percentage in a large real-world patient data set of clinical angiogram videos from UCSF, and externally validated it in a separate data set from the University of Ottawa Heart Institute.
The results of this study were published in JAMA Cardiology.
“If you have easy access to the coronary angiogram images, such as by directly accessing the storage service, it is easy to deploy CathEF or a similar deep neural network into clinical practice with minimal infrastructure changes,” Avram told Healio. “To this effect, the study team developed a custom software called PACS AI that leverages the storage service to allow doctors to apply CathEF — or similar algorithms — to coronary angiograms in near real time.”
A total of 4,042 angiograms from 3,679 adult patients were included in the analysis (mean age, 64 years; 65% men).
In the UCSF data set, CathEF discriminated reduced LVEF with an area under the receiver operating characteristic curve of 0.911 (95% CI, 0.887-0.934); a diagnostic OR for reduced LVEF of 22.7 (95% CI, 14-37); and, compared with transthoracic echocardiogram, a mean absolute error of predicting LVEF percentage of 8.5% (95% CI, 8.1-9), according to the study.
Avran and colleagues reported that although CathEF-predicted LVEF percentage differed by approximately 5% or less compared with transthoracic echocardiogram in 38% of data sets, differences of more than 15% were observed in approximately 15.2% of data sets.
In addition, the CathEF tended to overestimate low LVEFs and underestimate high LVEFs, according to the study.
External validation of AI for LVEF prediction
In external validation using data sets form the University of Ottawa Heart Institute, the researchers reported that CathEF discriminated reduced LVEF with an area under the receiver operating characteristic curve of 0.906 (95% CI, 0.881-0.931) and a mean absolute error of predicting LVEF percentage of 7% (95% CI, 6.6-7.4).
Moreover, the ability of CathEF to discriminate reduced LVEF and predict LVEF percentage was consistent across sex, BMI, low estimated glomerular filtration rate, presence of ACS, obstructive CAD and LV hypertrophy, according to the study.
“The study team at the Montreal Heart Institute is currently conducting further validation in a real-world setting by deriving measurements using CathEF in patients presenting with acute coronary syndrome to estimate the LVEF,” Avram told Healio. “This research demonstrates the increasing role of AI in health care. However, it's crucial to remember that AI tools like CathEF are designed to assist, not replace, clinicians. The adoption of such tools should be carefully managed to ensure they contribute to improved patient outcomes, and their use should always be guided by expert human judgement.”
For more information:
Robert Avram MD, MSc, can be reached at robert.avram.md@gmail.com.
Collapse
Click Here to Manage Email Alerts