Artificial intelligence-assisted peri-operative echocardiography shows promise but requires validation in hemodynamically unstable ICU patients and real-time intra-operative workflows.
This letter highlights that while AI-assisted peri-operative echocardiography shows promise in stable patients, its applicability in high-risk, hemodynamically unstable patients requiring real-time assessment remains a critical gap.
We read with interest the study by Borde et al. evaluating the diagnostic agreement of artificial intelligence-assisted peri-operative echocardiography in a multicentre cohort 1. The authors report good-to-excellent concordance between artificial intelligence and clinicians across 10 echocardiographic parameters, which represents a promising step towards integrating new technology into peri-operative practice. However, before widespread adoption can be recommended, we wish to highlight several considerations regarding generalisability and methodology that warrant discussion. Borde et al. did not study patients admitted to the ICU who required mechanical ventilation and vasoactive support. Yet it is precisely this haemodynamically unstable population who would benefit the most. Current consensus guidelines emphasise that a core value of peri-operative ultrasound lies in its ability to provide immediate assessment of high-risk patients 2. By not studying this population, the validation of the platform is effectively limited to elective, stable, non-ICU patients. Whether these findings can be extrapolated to truly high-risk peri-operative scenarios remains unclear. The study workflow required acquisition of a predefined 12-view sequence uploaded to a cloud platform, with the authors noting that artificial intelligence typically generates a report within 2 min. However, for intra-operative acute events, anaesthetists typically need to make intervention decisions within seconds. Acquiring 12 standard views itself takes several minutes, and when combined with uploading and cloud processing, a total duration of 5–10 min is a realistic estimate. During this interval, the patient's haemodynamic status may have changed substantially. The value of offline analysis may not align fully with the requirements of intra-operative real-time dynamic assessment. Yu et al. developed a novel method combining transoesophageal echocardiography with deep learning that automatically measures mitral annular plane systolic excursion 3. The system automatically acquires measurements every 5 min, and serial measurements correlated significantly with N-terminal pro-brain natriuretic peptide and high-sensitivity troponin T. Another study confirmed that this tool has excellent trend-tracking capability and is more precise than manual assessment 4. Whether similar artificial intelligence-driven continuous monitoring can be applied reliably to the transthoracic approach across different surgical populations remains to be established. Real-time video-stream analysis with immediate feedback would undoubtedly represent a more transformative advance than offline workflows alone.
“I cannot spend 45 minutes doing a preoperative scan. I have got only 5 minutes, maximum 10 minutes, to do my stuff and arrange things in my mind simultaneously. By the time the cart-based machine starts booting in, you are done with a point-of-care device.”
Zhang et al. (Fri,) conducted a letter in Peri-operative care. Artificial intelligence-assisted peri-operative echocardiography was evaluated. Artificial intelligence-assisted peri-operative echocardiography shows promise but requires validation in hemodynamically unstable ICU patients and real-time intra-operative workflows.