AI-aided echocardiographic quantification improves efficiency across operator experience levels: a prospective clinical trial

Background Echocardiographic (echo) assessment requires multiple measurements, is time-consuming and operator-dependent. Fully automated, artificial intelligence (AI)-based quantification has the potential to standardise analysis and improve workflow efficiency. Purpose To assess whether fully automated AI quantification reduces analysis time compared with manual assessment in clinical practice and across different levels of operator experience. Methods The prospective CORTEX-ECHO trial (Cost-, Resource-, and Time-Effectiveness Examination of Fully Automated Analysis in Echocardiography; German Clinical Trials Registry: DRKS00037757) enrolled 111 in- and outpatients (age 61±14 years; 45% female) who underwent a diagnostic echo-examination at our tertiary care centre. Left ventricular (LV) geometry and systolic and diastolic function were independently analysed by a fully automated AI tool and by two operators (beginner, expert), all blinded to each other’s results. The operators subsequently reviewed and, if necessary, adjusted AI-derived measurements. The primary endpoint was time required for manual analysis versus AI review and adjustment. LV geometry included end-diastolic/-systolic diameters and wall thickness. Systolic function was assessed by biplane Simpson ejection fraction (EF). Diastolic function comprised left atrial volume (LAV), transmitral E and A-wave velocity, septal and lateral e′ (tissue Doppler), and tricuspid regurgitation velocity (Figure 1). Measurements were performed on three consecutive heart beats. The secondary endpoint was inter-operator agreement for manual and AI-aided measurements using intraclass correlation coefficient (ICC) and mean absolute difference (MAD). Results AI-aided analysis reduced measurement time by 58% for the beginner and 56% for the expert (from 435±92 s to 184±58 s and from 316±73 s to 139±33 s, respectively) (Figure 1A). AI-aided analysis substantially reduced measurement time across all parameters in both beginners and experts, with the greatest time savings observed for biplane EF and diastolic assessment (Figure 1B-C). In beginners, AI reduced analysis time for LV diameters (86±31 s vs. 55±23 s), EF (197±53 s vs. 82±32 s), and diastolic assessment (152±29 s vs. 48±22 s). Comparable reductions were observed in experts for EF (170±45 s vs. 50±16 s), diastolic parameters (105±39 s vs. 60±20 s), and LV diameters (61±25 s vs. 28±17 s). AI reduced experience-related differences. For EF, agreement between beginner and expert improved from good with manual assessment (ICC 0.81; MAD 6.6%) to excellent with AI-aided analysis (ICC 0.97; MAD 2.2%). Agreement for LAV improved from excellent with manual assessment (ICC 0.93; MAD 11.5 ml) to near-perfect with AI (ICC 0.99; MAD 6.0 ml). Conclusion AI-aided Echo analysis significantly reduces analysis time and improves inter-operator agreement across experience levels, with the greatest efficiency gains observed in less experienced readers.