The automatic identification of left ventricular chambers and quantification of ejection fraction using a novel artificial intelligence-based system – a validation against cardiac magnetic resonance

Abstract

Abstract Background/Introduction Cardiac magnetic resonance imaging (CMR) is regarded as the reference method in assessing left ventricular (LV) ejection fraction (EF). However, 2-dimensional echocardiography (2D-Echo) is the most frequently used technique due to availability and practicability. The interpretation of 2D-Echo examinations depends on the user's expertise and may vary between different operators. A novel vendor-independent software based on artificial intelligence (AI) performs both, automated evaluation of 2D-Echo exams and calculations of LV EF in one workflow. Purpose We sought to assess the ability of the AI to automatically identify appropriate LV 4- and 2-chamber views (4CV) (2CV) from 2D-Echo exams and validate the resulting EF with CMR. Methods We consecutively enrolled 128 patients who underwent clinically indicated CMR examinations and performed a standard 2D-Echo at the same day. The server-based AI solution recognized the optimal LV 4CV and 2CV from 2D-Echo according to quality and depth criteria and automatically performed calculation of biplane EF by endocardial borderline detection. LV EF from CMR and AI were supervised by independent cardiologists blinded to the mutual results. Pearson's correlation (R) and Bland-Altman analysis with limits of agreement (LOA) were performed in order to assess bias between the two methods. Significance was defined as a 2-tailed P value <0.05. Results CMR was performed and LV EF was measured in all 128 patients. The median age was 60 years [20–86], 65% were males and CMR was performed due to coronary artery diseases (33%), suspected/florid myocarditis (20%) or further diagnosis of non-ischemic heart failure (47%). Eleven cases (9%) did not pass AI's criteria due to impaired acoustic window or poor 2D-Echo images. The AI system detected either 4CV or 2CV (ratio 1.2) in 13 patients (10%), and both 4CV and 2CV in 104 patients (81% overall feasibility) with a correct classification of 100%. For these 104 patients, excellent correlation was found for AI's biplane LV EF and LV EF from CMR with r=0.91 (p<0.001) (Figure 1, left). However, the absolute mean bias between AI and CMR was 3.5% (p<0.001) and LOAs were −10.6 and +17.5% (Figure 1, right). Conclusion The results provided by the AI-based software showed good capabilities and a perfect classification rate to identify 4CV and 2CV. In addition, the LV EF results were excellent compared to CMR, especially since our study did not include “echocardiographically” pre-selected patients. However, differences between AI and CMR measurements are not negligible and warrant further investigation. Funding Acknowledgement Type of funding sources: None.