Detection of Acute Myocardial Infarction from serial ECG using multilayer support vector machine

Acute Myocardial Infarction (AMI) remains a leading cause of mortality in the United States. Finding accurate and cost effective solutions for AMI diagnosis in Emergency Departments (ED) is vital. Consecutive, or serial, ECGs, taken minutes apart, have the potential to improve detection of AMI in patients presented to ED with symptoms of chest pain. By transforming the ECG into 3 dimensions (3D), computing 3D ECG markers, and processing marker variations, as extracted from serial ECG, more information can be gleaned about cardiac electrical activity. We aimed at improving AMI diagnostic accuracy relative to that of expert cardiologists. We utilized support vector machines in a multilayer network, optimized via a genetic algorithm search. We report a mean sensitivity of 86.82%±4.23% and specificity of 91.05%±2.10% on randomized subsets from a master set of 201 patients. Serial ECG processing using the proposed algorithm shows promise in improving AMI diagnosis in Emergency Department settings.