A Classification Tree Approach for Cardiac Ischemia Detection Using Spatiotemporal Information From Three Standard ECG Leads

The accurate noninvasive diagnosis of cardiac ischemia remains a great challenge. To this end, the ECG is the main source of information, and personal health systems may now embed intelligence for enabling any citizen to self-record an ECG anywhere at any time. Our objective is to find a decision-support approach that makes best use of these resources. A new classification tree based on conditions combinations competition (T-3C) is proposed for building a multibranch tree of combined decision rules, and its performance is compared to usual methods based either on discriminant analysis or on classification trees. Moreover, we assessed with these methods, the diagnosis content for ischemia detection of the spatiotemporal ECG information that can be retrieved either from the standard 12-lead ECG or from only the three orthogonal leads subset (I, II, and V2), easy to set-up in self-care. The diagnostic accuracy of 14 decision-making strategies was compared for ischemia detection induced by angioplasty on a test set from a study population of 90 patients. The best performance is obtained with the T-3C algorithm on three-lead ECG, reaching 98% of sensitivity and of specificity, thus exceeding 23% of the diagnostic accuracy of the recommended and currently used standard ECG criteria.