Enhanced acute myocardial infarction detection algorithm using local and global signal morphology

One shortcoming of conventional AMI detectors based on local morphologic features is that more subtle, globally distributed ECG changes (from the start of the QRS complex to the end of the T-wave) remain undetected. To characterize these changes, we develop two separate sets of basis vectors which span the subspaces occupied by the nonAMI ECGs and the AMI ECGs, respectively. The maximum likelihood estimate of the signal subspace is derived using the additive Gaussian noise model. A feature vector is computed by projecting the patient's ECG signal vector onto each of the basis vectors. A classification algorithm based on these global feature vectors performs significantly better than the conventional algorithm. Additional improvement is obtained by combining results from an optimized classifier using conventional local morphological measurements with the global feature classifier output to yield a combined decision. Test performance resulting from the local / global algorithm is Sensitivity 55% and Specificity 98% on a database of 1220 ECGs. A conventional ECG interpretive algorithm using localized ST-elevation and a rule-based classifier has Sensitivity 35% and Specificity 98%.