Top-resolution frequency analysis of electrocardiogram with adaptive frequency determination. Identification of late potentials in patients with coronary artery disease.

Abstract

Frequency analysis of the electrocardiogram with Fourier transform is a sensitive method of detecting late potentials. However, information about localization of late potentials is lost, frequency resolution is poor, and window functions have to be applied. We therefore analyzed multiple segments (25 msec long) of the surface electrocardiogram ("spectrotemporal mapping") with adaptive frequency determination (AFD), an autoregressive algorithm that is characterized by high-frequency resolution in very short segments without the use of window functions. Results were compared with those from Fourier transform and the Simson method. We studied 38 patients after myocardial infarction (MI) with sustained ventricular tachycardia (VT), 21 patients after MI without VT, and 18 healthy subjects. Frequency peaks could be clearly differentiated until a minimal interval of 6 Hz; with fast Fourier transform (Blackman Harris window) in a much longer segment (80 msec), the spectral peaks merged into one another at an interval of about 30 Hz. AFD revealed high-frequency components as narrow peaks in the range of 40-160 Hz in 28 of 38 patients (74%) after MI with VT. Because of the short segment size, exact localization of late potentials was possible; in most of the patients, the peaks occurred in segments inside the QRS complex and ended 20 +/- 10 msec after termination of the QRS complex. In patients after MI without VT, only four of 21 patients (19%) had spectral peaks in segments after the end of the QRS complex; however, 13 of 21 patients demonstrated microvolt potentials in segments within the QRS complex. These potentials did not extend beyond the end of normal ventricular activation. Only two of 18 healthy subjects showed abnormal AFD results. Patients with bundle branch block did not need to be excluded. AFD allowed good differentiation between late potentials and noise by a characteristic pattern of the spectral peaks. For the Simson method, patients with bundle branch block had to be excluded, and overall sensitivity was 42%. In five cases, the cause of failure of the Simson method could be identified as incorrect determination of the QRS limits due to noise. Thus, AFD is a promising method for detailed analysis of late potentials; it combines the advantages of frequency analysis (good differentiation between signal and noise and high-pass filters not necessary) and time domain analysis (localization of late potentials).