Adapting ECG morphology changes from reduced-lead set by specifically trained algorithms for acute ischemia detection

The reduced lead set in this study consists of a subset of 12-lead set with all the same limb leads, plus v1 and v5 chest leads. They were then converted to 12-lead signals with a trained matrix. The aims of this study are to evaluate the changes of morphology due to the conversion and the possibility of adapting the changes with the new criteria trained with a large ischemia ECG database. For the evaluation, we first applied the standard 12 lead computerized algorithm to the converted 12-lead ECGs. We then applied 2 training algorithms to the converted 12-lead ECG to obtain the specific criteria. One algorithm is based on neural network models, and the other is based on the binary-tree classification model with automatically induced rules (C4.5). The study used the chest pain databases from Mayo Clinic and Medical College of Wisconsin, which include ECGs from 993 anterior MI patients, 1187 inferior MI patients and 1987 Nonischemic patients based on their final confirmation with Cathlab or Biomarker tests. The results show that, without retraining, the reduced-lead set achieved 53% sensitivity for Inferior MI, compared with 54% from standard 12-lead set, and with both specificity of 99%. For anterior MI, the sensitivity of reduced lead set is 37%, compared to the 46% for standard 12 lead. After retraining, the reduced lead set has a sensitivity of anterior MI 56%, compared to a retrained standard 12 lead set of 57%, with specificity above 98% for both lead sets.