Feasibility of Artificial Intelligence–Based Electrocardiography Analysis for the Prediction of Obstructive Coronary Artery Disease in Patients With Stable Angina: Validation Study

Figure 1. Artificial intelligence (AI)–based quantitative electrocardiography (QCG) analysis. The diagram summarizes the study flow. The QCG analyzer transforms image-based electrocardiography (ECG) data into vectorized signal features. To validate the original QCG analyzer in patients with stable angina, we derived 5 QCG scores related to coronary artery disease (CAD) and evaluated their predictive value for the presence of obstructive CAD. Additionally, we developed a new QCG score that is more suitable for identifying obstructive CAD in stable angina by using a tree-based AI algorithm, XGboost, with the extracted vectors as input features. ACS: acute coronary syndrome; AUROC: area under the receiver operating characteristic curve; CNN: convolutional neural network; LV: left ventricular; STEMI: ST-elevation myocardial infarction.

Figure 2. Distribution of the quantitative electrocardiography (QCG) scores. (A) Among the 5 QCG scores, those for acute coronary syndrome (ACS) and myocardial injury exhibited the highest values. (B) When the QCG scores were stratified by the number of vessels with obstructive coronary artery disease (CAD), patients group with 3-vessel disease showed significantly higher scores for ACS and myocardial injury than those in the other groups. LV: left ventricular; STEMI: ST-elevation myocardial infarction.

Figure 3. Incremental predictive value of the original quantitative electrocardiography (QCG) scores for obstructive coronary artery disease (CAD). (A) The QCG scores for acute coronary syndrome (ACS) and myocardial injury demonstrated significant predictive performance for obstructive CAD, despite lower performance than that of clinical features. (B) In contrast, both scores showed moderate predictive performance for 3-vessel disease, comparable to that for clinical features, and provide incremental predictive value in combination with clinical features. AUROC: area under the receiver operating characteristic curve.

Figure 4. Prediction performance of the new quantitative electrocardiography (QCG) score for any obstructive coronary artery disease (CAD). This figure summarizes the discrimination performance of the new QCG score in the training and test sets. After categorizing the patients by the optimal cutoff values of the new QCG score (score of 66 in both training and test sets), the accuracy for obstructive CAD was 90.8% and 75.7% in the training and test sets, respectively. AUROC: area under the receiver operating characteristic curve; NPV: negative predictive value; PPV: positive predictive power.

Figure 5. Incremental predictive value of the new quantitative electrocardiography (QCG) score for any obstructive coronary artery disease in the test set. In the test set, the new QCG score showed significantly higher predictive performance for obstructive CAD than that for clinical features, and provided incremental predictive value in combination with clinical features. AUROC: area under the receiver operating characteristic curve.