Originally presented at the DGK-Herztage 2024. https://doi.org/10.1007/s00392-024-02526-y
Introduction:
Artificial Intelligence-based 5-lead 3D-vectorcardiography (5L3DVCG-AI) is easy to use, quantifies the individual risk for cardiac pathologies in need for further diagnostic procedures and may, with good results in women, facilitate and align the cardiological diagnostic pathway for coronary heart disease.
Methods:
In this multicentre retro- and prospective study, recordings from 5L3DVCG-AI were externally validated against automated 12-lead ECG in 287 patients with and without cardiac pathologies. 5L3DVCG-AI derived 12-lead ECG (VCG-ECG) was reconstructed from 5L3DVCG-AI with an algorithm (hgh-1.1.23) and time intervals were derived from 5L3DVCG-AI. Two independent specialist cardiologists masked for 12-lead ECG results interpreted VCG-ECG qualitatively and quantitatively. The following variables were compared between 5L3DVCG-AI and 12-lead ECG: electric heart axis and rhythm, HR, and time intervals for P, PQ, QT, QTcB. Presence of cardiac pathology (CP) was categorised as exclusion of any CP (control), mild CP or overt CP by 2 independent cardiologists from clinical practice with a follow-up period of 16.2 ± 7.5 months. Diagnostic accuracy was assessed for ECG findings and abnormalities. Correlation between VCG-ECG and 12-lead ECG was calculated for electric heart axis (Spearman’s rank correlation coefficient). Agreement was tested with Bland-Altman analyses. The modified PROCAM-score was used for cardiovascular risk factor (CVRF) assessment.
Results:
Of 287 patients (m:w 62:38%, 55.9 ± 16.1 years) of mixed ethnicity and moderate CVRF (2.1 ± 1.2), 70% were controls, 21% had mild CP and 9% overt CP. Strong correlations were seen for HR and electric heart axis (r=0.97 and r=0.71, p<0.001 respectively) with 12-lead ECG which was visually confirmed. Quantification of ECG variables such as times for P, PQ, QT, QTcB showed strong correlations (all p<0.001), low systematic bias (SB; -0.9 to -3.9%) and narrow 95% upper and lower limits of agreement (uLoA; 5.6 to 17.3%, lLoA; -8.5 to 19.1%). In women, 5L3DVCG-AI at rest is strong in detecting CVR (r=0.71, p<0.001, mod. PROCAM-Score) and in differentiating cardiac pathologies (β=0.24, T=2.64, p<0.05, corrected for CVR). ECG abnormalities (AF, delayed R-progression, cardiac ischaemias, LVOT VES, sinus bradycardia, sinus tachycardia, AT, LAHB, LHH) were visually detected from 12-lead ECG and VCG-ECG with a sensitivity of 75% and specificity of 100% with low interrater variability. All clinically relevant ECG pathologies were displayed in both systems. The VCG-ECG had only small differences regarding the amplitude of the QRS-complex in three cases.
Conclusion:
In summary, 5L3DVCG-AI is an easy-to-use and feasible technology with good accuracy and reproducibility for electric heart axis, ECG-parameters and intervals and thus offers additional value in detecting individuals with cardiac pathologies or cardiac risks. 5L3DVCG-AI may replace conventional 12-lead ECG in the General Practice or cardiological outpatient departments. Especially for women this may offer additional value.
Zur Studie