Detection of inferior myocardial infarction: A comparison of various decision systems and learning algorithms

In this work we have focused on classification of inferior myocardial infarction (MI). We compared the best known scoring/coding/decision systems (the Selvester QRS score, the Novacode, and the Siemens 440/740) and several learning algorithms (Ripper, C4.5, and SVM). The decision systems were developed with different purposes (the Selvester for estimation of MI size, the Novacode for clinical and epidemiologic studies, and the Siemens for ECG device Siemens 440/740). In this work we combined these systems with additional simple rules and compared performance to: (i) decision systems alone, (ii) base classifiers (Ripper, C4.5, and SVM). Our database consisted of 2596 ECG records annotated by experienced cardiologist. Among decision systems the Selvester and the Siemens had F-measure 54% and 51%, respectively. Meaning that about 50% of MI´s were correctly classified. Even lower F-measure of 39% was obtained by Novacode. Better results were achieved using rule miner Ripper with F-measure of 68%, however, due to a number of rules created, the resulting model was hard to interpret. Last, combination of decision systems with additional simple rules created by AdaBoost yielded the best performance with F-measure 71%, sensitivity (Se) 78%, and specificity (Sp) 95%.