Evaluation of ventricular late potentials in the frequency domain at various bandwidths

DA Balderson, DWG Harron, RH Mitchell, OJ Escalona

Research output: Contribution to journalArticlepeer-review


IntroductionVentricular late potentials (LP) have been demonstrated in subjects prone to a major or fatal arrhythmic event (Breitirardt et al., 1981). However, various results have been reported with different methods of analysis (time and frequency domain). Furthermore, within each method, the Parameters and endpoints used have been different (Haberl et al., 1988). The present study attempted to differentiate between the frequency bands for subjects with and without LP who had previously been identified using the time domain method. MethodsHigh resolution surface ECGs were recorded from 82 subjects using the XYZ orthogonal lead system and signal averaged (‘Predictor'-Corazonix). In the time domain, 13 of these subjects were identified as LP positive. A Fourier transform (Blackman-Harris window) was performed on a 120 ms segment of each lead starting from the 40 μV level at the end of the QRS and extending out towards the T wave. Areas (A) and area ratios (AR) for the spectral plots for 11 different frequency bands (B1- B11) were calculated (25-45, 45-65, 65-85, 85-105, 105-120,25-65, 65-95, 95-120, 25-75, 75-120, 25-120 (Hz) for each lead using the A and AR data respectively (AR = A/total area of the spectrum (25-120 Hz)) (Harberl et al., 1988). Discriminant analysis (SPSS, Advancedstaristics V2.0) was performed for all subjects on both the A and AR data.ResultsThe results discriminated between subjects with and without LP and gave a range for total predictability (i.e. total percentage (%) of 'grouped' subjects correctly classified) across all leads (mean XYZ, Z, Y and Z) and all bands (B1-B11) of between 82.9 to 93.9o/o and 62.2 to 73.1% for the A and .AR data respectively. In both the A and AR data the lead for maximum total predictability was the X lead (93.9 and 73.1% respectively). The predictiveness (%) for correct group membership of normals and abnormals over all leads and bands gave ranges of 92.8 to 98.6%. and 30.8 to 62.9% for the A data compared with ranges of 63.8 to 75.4% and 53.8 to 69.2% for the AR data. The three most predictive bands for discriminating between subjects with and without LP in each Iead were also analysed. From this analysis, band 10 (75-120 Hz) for the A data of the X lead had the best predictiveness with an 85.5 and 61.59/o chance of subjects being grouped correctly as normal (no LP) or abnormal (with LP) respectively. Total predictability for this was 81.7%. For the AR data the most predictive band for correct grouping was band 1 (25-45/25-120 Hz) again in lead X which gave results of 59.4 and 62.9% for normal and abnormal grouping respectively with a total predictability of 60.9%.In conclusion, the results indicate that using discriminant analysis the A data gave the best prediction (p <0.05) using the X lead and a band of 75-120 Hz (B10, 81.7%) for discriminating between subjects with and without LP. In AR data the best predictor (NS) was lead X and band 1 ((25-45) 25-120 Hz, 60.9%). Overall the A data consistently gave increased predictability indicating that area ratio data in spectral analysis may be incorrect when comparing the results with those obtained using the time domain. However, it may also suggest that the parameters used to identify LP in the time domain may have to be reassessed.Breithardr. G. et al. (1981). Eur. Heart J.. 2. 1.Haberl. R. et al. (1988). J. Am. Coll. Cardiol., 12. 150.
Original languageEnglish
Pages (from-to)583-584
JournalBritish Journal of Clinical Pharmacology
Publication statusPublished (in print/issue) - 15 Sept 1991


  • ventricular late potentials
  • ECG frequency domain analysis
  • high-resolution ECG
  • electrocardiography
  • Cardiology
  • signal averaged ECG
  • discriminant analysis
  • abnormal LP.


Dive into the research topics of 'Evaluation of ventricular late potentials in the frequency domain at various bandwidths'. Together they form a unique fingerprint.

Cite this