A clustering based system for instant detection of cardiac abnormalities from compressed ECG

Compressed Electrocardiography (ECG) is being used in modern telecardiology applications for faster and efficient transmission. However, existing ECG diagnosis algorithms require the compressed ECG packets to be decompressed before diagnosis can be applied. This additional process of decompression before performing diagnosis for every ECG packet introduces undesirable delays, which can have severe impact on the longevity of the patient. In this paper, we first used an attribute selection method that selects only a few features from the compressed ECG. Then we used Expected Maximization (EM) clustering technique to create normal and abnormal ECG clusters. Twenty different segments (13 normal and 7 abnormal) of compressed ECG from a MIT-BIH subject were tested with 100% success using our model. Apart from automatic clustering of normal and abnormal compressed ECG segments, this paper presents an algorithm to identify initiation of abnormality. Therefore, emergency personnel can be contacted for rescue mission, within the earliest possible time. This innovative technique based on data mining of compressed ECGs attributes, enables faster identification of cardiac abnormalities resulting in an efficient telecardiology diagnosis system.