An editing method for computer-assisted ambulatory ECG review systems.

Abstract

Editing of computer-assisted ambulatory ECG reviews is critical for accuracy and quantification of the arrhythmias present. This may be time consuming for high arrhythmia content types or tapes with noise interference. A system that combines probit analysis and principle components transformation coupled with maximum likelihood decision theory, through identification of the complexes most subject to classification error in the initial review process and by correcting residual errors automatically, can decrease the number of complexes which need to be overread by a human editor. Probit analysis assigns a statistically derived value, a z value, to complexes classified as normal sinus or ventricular premature complexes. This permits the grouping of complexes for which the likelihood of being normal sinus or ventricular premature is high and a group which has a greater likelihood of being incorrectly classified. This latter group can be shown to the human editor for verification or correction of classification. The complexes undergo principle components transformation which describes the QRS by a set of derived components. When a classification is approved or changed by the human editor, the computer, utilizing maximum likelihood decision rules, moves ahead in the tape to correct the classification of the remaining unedited complexes on the basis of the similarity of their principle components profile to the edited ones. The system reduced total errors, false positive or false negative, to less than one percent in all of the high arrhythmia and noise content tapes used for this study.