Although the precision of automated machine learning (AML) for analysis of cardiac structures matches that of expert clinicians and medical trainees, AML systems are >100 times faster in measuring changes in left ventricular ejection fraction (LVEF) in several cardiovascular disease states, according to a study published in Circulation: Cardiovascular Imaging.

A total of 110 patients with myocardial infarction (n=32), left ventricular hypertrophy (n=17), cardiomyopathy (n=17), or other pathology (n=14) and healthy volunteers (n=30) underwent scan:rescan cardiovascular magnetic resonance imaging. An expert with >15 years’ experience, 2 trained junior clinicians, and a fully automated convolutional neural network trained on 599 multicenter disease cases measured LV chamber volumes, mass, and LVEF.

The minimal detectable change in LVEF as detected by expert analysis was 8.7%, or 20 g in LV mass. Scan:rescan precision for LVEF was similar for expert (coefficient of variation [CV], 6.1%; 95% CI, 5.2%-7.1%; P =.2581), trained junior (CV, 8.3%; 95% CI, 5.6%-10.3%; P =.3653), and automated analysis (CV, 8.8%; 95% CI, 6.1%-11.1%; P =.8620). The time it took for humans to analyze LV metrics was 13 minutes compared with 0.07 minutes for automated analysis, indicating that automated analysis is 186 times faster than human assessment.

Limitations of the study were the small number of observers to measure variability in precision, as well as the short scan-rescan interval (ie, 82% of studies were obtained on same day).

“Given that a major source of measurement variability is attributable to the observer,” the researchers wrote, “automated approaches offer the future potential to surpass human experts, demonstrable using this scan-rescan resource.”

Automated Machine Learning Precisely Measures Cardiac Structural Changes on CMR

Want to read more?

Want to read more?

Related Articles

Want to read more?

Want to read more?