Improved methods for combined measurements and analysis in the time- and frequency domain simultaneously

Why ?

The standard HiRes ECG with diagnostics in time-domain, as implemented in our computerized system (High-resolution signal averaged XYZ vector magnitudes) was shown by T. Buckingham rt al. (1989)  to have sensitivity of 62%, specificity of 75%, accuracy of 70%, a positive predictive value of 58%., and a negative predictive value of 78%. The relative low positive prediction accuracy in time domain emphasized the need for continued methodological and technical refinements. Spectral methods, as proposed by Dr. M. Cain et al (1984, 1985), did not yield a clinically applicable diagnostic method.

Click here for my more detailed comment on the problems [1,3] in the detection of late potentials.

Our purpose was to improve the diagnostic capability of the computer-based diagnostic system by introducing an additional, spectrum-derived parameter into the decisionmaking logic. This parameter was an energy Spectral Area Ratio SAR [3], calculated by dividing the spectral energy area from 20 to 48 Hz of the terminal QRS, by the area from 6 to 20 Hz. This ratio characterized the relative energy of the higher-frequency late potentials. In combination with the standard time domain parameters (QRS width, V40 and LAS), SAR yielded excellent results (R>1, QRS wide, and V40 or LAS fulfilled). In a study which included 10 sustained-VT patients and 14 non-VT patients (6 normal and 8 MI with no VT), we were able to distinguish all patients at VT risk using our decisionmaking logic. Click here to obtain scatter plots of the Spectral Area Ratio (SAR) against the three time-domain parameters for the VT (n=10) and non-VT patients (n=14) [3].
 

Using Time Frequency Distributions

The idea of combining time and frequency domain by using time-frequency representations was first seriously tested by P. Lander, D. Albert and Ed Berbari (1992). They studied time-frequency distributions during the QRS complex, which they called spectro-temporal analysis, or spectrotemporal mapping (STM). In an effort to increase temporal resolution, the authors chose a time interval of length T = 16 ms, which however appears too narrow to achieve satisfactory spectral resolution, although a higher sampling rate of 2000 Hz was used. As later Drs. Berbari and Lander (1992) point out, STFT has significant short-comings for signals of short duration with a time varing spectrum. The literature has a number of conflicting studies in evaluating spectral analysis (Wheeler et al., 1992, Blaszyk et al.,1992). Though Kelen (1992) reported an improved performance of STFT with regard to detecting LP's in the presence of bundle branch blocks, the problem of identifying VT risk in patients with conduction abnormalities still remained.

In a study and related paper [1] we examined the continuous Wavelet Transform (WT) as an alternative to classical Fourier Transform and STFT in the detection of LP's. Though WT suffers from the same time-frequency resolution limitations (it can not be made arbitrarily good simultaneously in both domains), it is more suitable for decomposition of signals containing high frequency components of short duration (Rioul, Vetterli, 1991).

You can click here to see the results obtained by applying the WT on two patients: Time frequency distributions (Wavelet Transform, WT) and Short-Time Fourier Transform (STFT) of the XYZ vector magnitude of the body-surface electrocardiogram of a patient with and a patient without ventricular tachycardia (VT).

In the example above, both WT and STFT performed well in separating VT from non-VT, with the WT being slightly better [1]. The WT has the advantage over STFT, that it has a higher resolution for signals of low-amplitude, short-duration and higher frequency - exactly of the late-potential type. With it, one doesn't have the difficulties with selecting window lenthg, shape and time overlap, typically encountered with the STFT. In addition, with the WT a variety of analyzing wavelets can be used, to enhance different transformation features [1].

Related Publications:

[1] Gramatikov, B., Georgiev, I.   Wavelets as an Alternative to STFT in signal-averaged electrocardiography.    Medical and Biological Engineering and Computing, Vol.33, No.3, May 1995, pp. 482-487.

[2] Nikolov, Z., Georgiev, I., Gramatikov, B., Daskalov, I.    Use of the Wavelet Transform for Time-Frequency Localization of Late Potentials.  Biomedizinische Technik, Suppl.38 (1993), pp. 87-89.

[3] Gramatikov, B.   Detection of late potentials in the signal-averaged ECG - combining time and frequency domain analysis.   Medical and Biological Engineering and Computing, Vol.31, No.4, July 1993, pp.333-339.

[4] Gramatikov, B.    Digital filters for the detection of late potentials.   Medical and Biological Engineering and Computing, Vol.31, No.4, July 1993, pp.416-420.

[5] Gramatikov, B.   Non-invasive recording of His-bundle activity by means of high-resolution signal-averaged electrocardiography.   Sixth National Conference on biomedical Physics and Engineering with international Participation. Sofia, 22-24 October 1992., (in English), In Proceedings, pp.3-8.

[6] Gramatikov, B.   Comparison of some linear phase FIR-filters for real-time ECG processing.   Sixth National Conference on biomedical Physics and Engineering with international Participation. Sofia, 22-24 October 1992 (in Engl.)   In Proceedings - pp.27-31.

[7] Shachov, B., Gramatikov, B., Petkov, A., Dimitrov, E.    Late potentials in patients with ventricular tachychardia - initial successful clinical experience with the new technology of CLEMA.   Fourth National Congress in Cardiology. Plovdiv, Bulgaria, June 4-5, 1992. In Abstracts - p.28, Abstract 62. ( in Bulg.).

[8] Gramatikov, B.   Methods of generating and testing of digital filters for biological signals (Bulg.)   Fourth national scientific session Automation in biotechnological processes and biomedical research, Sofia, 17-19 Sept. 1991; in Proceedings, pp. 210-217.