Automated Measurements of Left Ventricular Ejection Fraction During Cardiac Surgery by Transesophageal...

There is limited literature on use of artificial intelligence for systolic function assessment in cardiac surgery with transesophageal echocardiography (TEE). The present study aimed to evaluate the correlation between conventionally measured ejection fraction (EF) by clinicians and the automated measured EF (Auto-EF) with TEE in patients undergoing cardiac surgery.

Muti centre, retrospective, observational, diagnostic accuracy study.

Four tertiary care level hospitals in India.

180 adult patients undergoing various cardiac surgical procedures.

None.

The manual EF was calculated using the biplane method of disks. After the manual measurement, the "Auto-EF" measurement was performed with "Echo PAC" (GE HealthCare, Chicago, IL, USA) version 206.The interclass correlation for EF was 0.84 (95% CI 0.68-0.91), indicating significant agreement (p < 0.001) between manual and auto EF measurements. Pearson's correlation coefficient (r) was 0.95 (95% CI 0.94-0.97) for end-diastolic volume; 0.94 (95% CI 0.93-0.96) for end-systolic volume, and 0.88 (95% CI 0.84-0.91) for EF (all p < 0.001) between manual and auto measurements. The mean difference between manual and auto EF was 3.30%, with limits of agreement from -8.03 to 14.64% according to Bland-Altman analysis. There was a statistically significant agreement between the auto and manual methods for EF grade (p < 0.001), with Cohen's kappa of 0.72 (95% CI 0.64-0.79). The auto method correctly classified 72.2% of cases compared to the manual method, while 23.3% were over-classified and only 4.4% under-classified.

The TEE measurement of manual and auto EF in cardiac surgical patients showed a strong agreement, with only a slight difference in average measurement. This method has the potential to save time compared to manual endocardial border tracing, enhance measurement consistency, and can be recommended for clinical use during cardiac surgery.