Classification Algorithm for Complexes in Long Term ECG Data Records

Long term ECG recording is a key tool in modern clinical practice to achieve high-efficient diagnostic accuracy. The manual approach requires extensive efforts by qualified medical personnel and thus enhances drastically the cost of clinical investigation. In addition, it may fail due to high noise levels and various artifacts in the recorded ECG. The paper discusses an algorithm to classify ECG complexes in a long term recording obtained during Holter monitoring to present them in a user-friendly form to a medical doctor. The algorithm also includes the procedure of averaging those complexes that appear close to each other to increase the signal-to-noise ratio. The classification criteria include heart rate, cross-correlation coefficient, ECG complex energy level. The heart rate estimation procedure is applied to select those complexes with close heartbeat intervals. The correlation coefficients are used to select those complexes which are rather identical in their shape pattern. In addition, to avoid averaging among ECG complexes with amplitudes at about 10-fold difference, they are also selected according to their energy level. The suggested approach aims both at improvement of the diagnostic procedure by automatic online analysis and reducing the diagnostics costs.