Luu, D.H.L. (2026) Artificial intelligence–derived versus cardiologist left ventricular ejection fraction agreement in novice-acquired point-of-care ultrasound in critically ill patients: a prospective study. thesis, Medicine.
Full text available on request.Abstract
Background: Artificial intelligence (AI)-based algorithms for automated left ventricular ejection fraction (LVEF) assessment show good accuracy in stable patient populations, but their reliability in critically ill patients undergoing novice-acquired point-of-care ultrasound (POCUS) is uncertain. Objective: To assess agreement between AI-derived LVEF from images acquired by a trained novice operator and cardiologist interpretation of the same images in intensive care unit patients. Methods: In this prospective study, a trained novice operator acquired apical four-chamber images during clinically indicated POCUS in 34 adult ICU patients. One cardiologist, blinded to the AI results, reviewed the images. Agreement was assessed using an absolute-agreement intraclass correlation coefficient (ICC), Bland�Altman analysis and coverage probability (CP). Results: Paired measurements were available for 27 patients (79.4%). Median AI LVEF was 53.0 (IQR, 48.0% - 59.0%) while median cardiologist LVEF was 55.0% (IQR, 50.0% - 60.0%). The ICC was 0.483 (95% CI, 0.127– 0.727; P=0.005). Mean signed difference was −0.6 percentage points, mean absolute difference was 8.1 ± 4.8 percentage points, and limits of agreement were −19.4 to +18.2 percentage points. CP within ±10 percentage points was 0.741 (20/27). Green-quality examinations showed higher agreement with an ICC of 0.760 (95% CI, 0.451–0.908) than yellow-quality examinations with an ICC of -0.562 (95% CI, −0.709–0.164). Conclusion: AI�derived LVEF showed minimal average bias but limited overall agreement and substantial individual-level error. It should not yet replace expert assessment for individual ICU patients. Green image-quality feedback seems to be necessary for good agreement but requires validation in larger studies.
| Item Type: | Thesis (UNSPECIFIED) |
|---|---|
| Supervisor name: | Grady, Dr. B.P.X. |
| Faculty: | Medical Sciences |
| Date Deposited: | 20 Oct 2026 11:33 |
| Last Modified: | 20 Oct 2026 11:33 |
| URI: | https://umcg.studenttheses.ub.rug.nl/id/eprint/3979 |
Actions (login required)
 |
View Item |