Artikel
Wavelet or ARMA models for automatic detection of B-waves?
Was ist besser für eine automatisierte B-Wellen-Erkennung: Wavelets oder eine ARMA-Modellierung?
Suche in Medline nach
Autoren
-
M. Kiefer
- Neurochirurgische Klinik, Universität des Saarlandes, Medizinische Fakultät, Homburg-Saar
-
S. Jetzki
- Lehrstuhl für Medizinische Informationstechnik, Helmholtz-Institut für Biomedizinische Technik, RWTH Aachen
-
S. Szczesny
- Lehrstuhl für Medizinische Informationstechnik, Helmholtz-Institut für Biomedizinische Technik, RWTH Aachen
-
W.I. Steudel
- Neurochirurgische Klinik, Universität des Saarlandes, Medizinische Fakultät, Homburg-Saar
-
S. Leonhardt
- Lehrstuhl für Medizinische Informationstechnik, Helmholtz-Institut für Biomedizinische Technik, RWTH Aachen
| Veröffentlicht: | 8. Mai 2006 |
|---|
Gliederung
Text
Objective: To find an optimized model for computer assisted analysis of intracranial pressure (ICP) monitoring.
Methods: ICP was monitored using a special PC-ICP software tool and a Spiegelberg III probe in 100 patients suffering from various types of hydrocephalus. The ICP levels were sampled and recorded at 1 Hz A LabVIEW® software tool was developed to find episodes of A- and B-wave activity within the ICP data using an ARMA model and various other wavelet models. Additionally, an experienced neurosurgeon determined whether there was B-wave activity within the identical ICP data. The results from both the human and computer analyses were then compared. The neurosurgeon was not aware of the algorithms’ analyses of the ICP data.
Results: Using the given parameters from the ARMA or wavelet calculations, the maximum concordance of the computer analysis to the human expert’s decisions was about 80%. Compared to the other wavelets, the ARMA model was more sensitive to B-waves, even for an artefact-loaded ICP signal. The wavelet analysis provided more precise results in locating and seperating B-waves, but it characterized at most 70% of the B-waves found by the expert’s analysis. Nevertheless, 90% of the B-waves noted through the wavelet model was confirmed by the human expert, whereas the ARMA model may recognize B-waves that can not be corroborated by the expert because of an artefact overload in the signal. The major advantage of ARMA and wavelet models for an automated ICP analysis is that both allow a more precise recognition of the B-waves’ amplitudes in contrast to the previously used FFT analysis. The FFT requires a periodical signal of at least 4 periods, where the ARMA model requires only two and the wavelet model recognizes even single waves. This is important if B-wave analysis should be reconsidered for shunt indication.
Conclusions: Automated A- and B-wave detection with an ARMA or a wavelet model seems promising. However, further efforts are necessary to optimize the parameterization of both models.
