According to the requirement of the computer-aid electrocardiogram (ECG) automatic analysis, we concluded the most important features concerned by physcians. They are peak features, interval features, baseline features and ST morphology features. In this paper, we introduced the detection process of ECG peak and boundary features with the traditional threshold based method and the detection method of ST morphology. We paid more attention to our experience based multi-lead decision model on ECG wave boundary detection.These detection algorithms were assessed in actual ECG data. According to physcians'feedback, we found that:a) the assessment accuracy of peak features is high; b) Interval features'detection what also means wave boundaries'detection is a classical problem in ECG automatic analysis. But we got the total accuracy of 92.4% by combining the method of single lead's extraction and experience based multi-lead decision model; c) The ST morphology feature has high relevance to the baseline feature. So when baseline draft happened inside a complete ECG beat, the assessment result would not be so accurate. This needs to be improved.
|