Novel ECG features and machine learning to optimize culprit lesion detection in patients with suspected acute coronary syndrome