Examining R Peak Changes in Ischemic Condition using Random Forest Classifier under Stress Test ECG

Abstract

The most common clinical criteria for electrocardiographic diagnosis of myocardial ischemia include ST segment changes but not the QRS morphology. This study aimed to interpret the variations in the amplitude of the R wave that might be helpful in better diagnosis of ischemia during stress testing along with ST-segment changes. Exercise Stress Test (EST) electrocardiograms (ECGs) were reviewed in 152 subjects (89 ischemic and 63 non-ischemic) including 51 females and 101 males of mean age (56 ± 4.24 years). R wave amplitude was measured in the supine and immediate post-exercise period. ST segment variations were also measured during load and recovery stages at J+ 80ms. An index known as ΔRST was formed from the summation of variations in the amplitude of the R wave and average ST segment changes. At a cutoff point of -1.5 mm ΔRST index provided a sensitivity of 71.99% and a specificity of 63.63% in contrast to the ST segment criteria that showed 55.9 % sensitivity and 54.54% specificity due to a large number of false positive and negative responses. The Random Forest (RF) model showed the best performance for classifying ischemic and non-ischemic conditions using ΔRST criteria for both lead II and lead V5. The accuracy, precision, and F1 score of the model for lead II were 0.96, 0.94, and 0.97 respectively. The performance matrices for lead V5 are 0.96, 0.98, and 0.94 respectively.