Abstract Background Echocardiography is widely used for identification of wall motion abnormalities in patients with suspected coronary artery disease. Segmental wall motion (SWM) is evaluated in practice, qualitatively, by visual estimation. SWM evaluation is known to be complex and time consuming, demands high level of expertise and subjective, especially when providing segmental scores to individual segments. To assess regional rather than global impairment, individual segments, when identified as abnormally contracting, are further evaluated to assess if they are linked to a muscle territory supplied by a specific coronary artery, which may imply impairment in blood supply. We labeled this decision-making process as "Territory logic" (TLG). The Left Anterior Descending (LAD) coronary artery supplies blood to a large mass of the left ventricular muscle, thus, when involved, proximal obstruction is life threatening and early diagnosis may be crucial. An artificial intelligence (AI) based platform for automated analysis of echocardiographic examination includes a machine learning based module, trained using 735 echo exams, for automated SWM evaluation. The module utilizes the four, two and three chamber apical views and provides segmental scores based on 16 segments model, scoring each segment as normal, hypokinetic or akinetic. Purpose To evaluate the ability of an AI based application to identify contractility impairment in the LAD territory compared to visual estimation, when applying TLG to both methods. Methods A total of 161 exams collected from two medical centers were included in the study, 64% male, mean age 60±16 years. 48 (29%) had normal LV function, 76 (47%) had previous myocardial infarction. The SWM evaluation by visual estimation was used as reference. Four, two and three chamber apical clips were manually selected from each exam. The module processed the clips automatically. The identification of impairment in the LAD was evaluated by applying TLG to both methods (Fig.1). Results 143 exams (89%) in which automated processing was possible for all three apical views were included in the analysis. The comparison between the automated and reference results showed a good ability to identify impairment in the LAD by TLG, with specificity and sensitivity of 87% and 80% respectively. The results are summarized in (Table-1). Conclusions The AI based platform for automated SWM evaluation can assist in identifying wall motion impairment in the LAD territory, comparable to the logic used in clinical practice.
Read full abstract