Artificial Intelligence Versus Anesthesiologist Visual Assessment of Ejection Fraction by Transesophageal Echocardiography During Cardiac Surgery

This was a prospective, single-cohort observational study conducted at the Cardiothoracic Surgery Academy, Ain Shams University. Adults aged 21 years or older undergoing elective coronary artery bypass grafting or valve repair or replacement surgery were enrolled consecutively after providing written informed consent.

After induction of anesthesia and hemodynamic stabilization, transesophageal echocardiography images were obtained using a transgastric short-axis view at the mid-papillary level and an angle of 0 degrees. End-diastolic and end-systolic frames and an M-mode recording were collected for assessment of left ventricular ejection fraction.

Pseudonymized echocardiographic images were assessed using a general-purpose artificial intelligence application. The same images were independently assessed by two blinded professor anesthesiologists who had no access to the artificial intelligence output or the other assessments. Visual estimation of ejection fraction, M-mode-derived ejection fraction, and interpretation time were recorded. These findings were also compared with the preoperative transthoracic echocardiography assessment.

The artificial intelligence output was used exclusively for research and was not used to make or modify any clinical decision. A total of 93 eligible participants were included in the final analysis.

Inclusion criteria

• Age 21 years or older.
• Scheduled for elective cardiac surgery involving coronary artery bypass grafting, valve repair, or valve replacement.
• Availability of a preoperative transthoracic echocardiographic assessment.
• Either sex.
• American Society of Anesthesiologists physical status III or IV.

Exclusion criteria

• Coagulopathy, defined as an international normalized ratio greater than 2 or a platelet count below 50,000/mm3.
• Previous cardiac surgery (redo procedure).
• Esophageal or upper gastrointestinal pathology preventing safe transesophageal echocardiography probe placement, including esophageal stricture, tumor, or large varices.
• Poor-quality transesophageal echocardiography images despite standard optimization, preventing reliable interpretation.
• Severe arrhythmia or hemodynamic instability before surgery, during induction, or during surgery that could compromise the accuracy of the transesophageal echocardiography assessment.
• Refusal to participate or a condition limiting valid consent or participation, including cognitive impairment or a language barrier.