Morphologic features of the ECG for detection of stress-induced ischemia

In this work we evaluated the improvement in the detection of stress induced ischemia achieved by ECG features not commonly measured by cardiologists. We evaluated 927 patients recorded during treadmill exercise SPECT following the Bruce protocol. The patients were labeled in three groups by two experts according to their SPECT images as: no ischemia (Ischemic Miocardium (IM)< 5%), mild ischemia (5% ≤IM< 10%) and moderate/severe ischemia (IM≥ 10%). The features studied were grouped as pretest features, exercise features and those proposed in this work. Among the features proposed, we included the Q, R, and S wave amplitudes, the slopes around each wave, and estimates of the P and T wave widths, as well as the QRS complex width. We also studied other features related to the ST segment. For the detection task, we used logistic regression, while the performance was evaluated using k-fold crossvalidation. Most discriminant features were selected using a feature selection algorithm for two thresholds of ischemia, ≥ 5% and ≥ 10%. Finally, two models of features were evaluated and compared to those reported by Sharir et al. obtaining improved performance for IM ≥ 5% in Se 77%, Sp 88% and AUC 0.88; while for IM ≥ 10%, Se 76%, Sp 90% and AUC 0.95.