Enhancement of Stress ECG Performance with Machine Learning

Figures

• 

  Figure 1. Model Overview Data collection, training, validation, and testing methodology. AI = artificial intelligence; CAD = coronary artery disease; CF = clinical factors (gender, age, risk, blood pressure, so on); ESE = exercise stress electrocardiogram.

• 

  Figure 2. Artificial Intelligence Model vs Human Performance (A) Receiver operating characteristic curve for expert knowledge guided transformer architecture. Comparison of model performance with physician interpretation in (B) accuracy, (C) negative predictive value (NPV), and (D) sensitivity. P value for McNemar test comparing paired proportions. A bootstrap approach was used to calculate the 95% CIs for the difference in performance between models. Error bars represent the 95% CIs for the reported measure. Abbreviations as in Figure 1.

• 

  Figure 3. Model Performance by Sex Model performance between males and females in (A) negative predictive value (NPV), (B) positive predictive value (PPV), (C) accuracy, (D) sensitivity, and (E) specificity. P value for two-sample test for equality of proportions with continuity correction comparing independent groups. Error bars represent the 95% CIs for the reported measure.

• 

  Central Illustration. Artificial Intelligence–Based Model to Detect Obstructive Coronary Artery Disease Using single-center data, we developed and tested an artificial intelligence model capable of extracting and analyzing ECG waveforms from exercise stress tests to predict obstructive coronary artery disease (defined as >70% stenosis) on coronary angiography. The model demonstrated comparable accuracy between females and males and outperformed human interpretation. External validation confirmed favorable performance, with a specificity of 93% and a sensitivity of 65%. Error bars represent the 95% CIs for the reported measure. AI = artificial intelligence; ECG = electrocardiogram.