Intelligent electrocardiogram pattern classification and recognition using low‐cost cardio‐care system

Abstract

Electrocardiogram (ECG) contains detailed information regarding incidental abnormality of a subject. Manual analysis of a long time ECG record is a lengthy process. Computerised ECG analysis supports clinicians in decision making. While designing a low-cost diagnostic support system, constraints on the system resources limit the processing speed, eventually affecting the reliability. To resolve these issues, three key factors have been addressed in this study: the feature extraction method, total number of features and the database used. For feature extraction, ‘polar Teager energy’ algorithm has been developed, yielding nearly 70% saving in processing time as compared to other well-known methods. Using features with linear relationship leads to reduction in feature vector dimension, without compromising its classification performance. Therefore the linear relationship between two ECG features, namely ‘informational entropy’(S) and ‘mean Teager energy’ has been revealed. These features are utilised for ECG beat classification using ‘fuzzy C-means clustering’ algorithm. The algorithm is evaluated using the MIT-BIH database and then tested by ECG measured with the cardio-care unit. The QRS detection performance of the proposed method is very good, with 0.27% detection error rate. For classification of ECG beats, average sensitivity and positive prediction rate achieved are 98.93% each.