## Page 49

# proefschrift gommer inhoud met kaft.indd

Dynamic cerebral autoregulation in CVR testing
Except for 1 patient all were on statin treatment and about 50% of the patients
were on some kind of antihypertensive treatment. Medications were not discon-
tinued during the testing phase.
The measurements were performed in the non-acute stroke phase, i.e. 6 weeks to
6 months post stroke (median 93, ranging from 50 to 190 days after stroke).
Data analysis
An automatic algorithm in the Task Force Monitor detected R-waves from the
ECG and between every two R-waves the diastolic, systolic and mean values of
the blood pressure waveform was determined. Correspondingly the peak-
systolic, end-diastolic and mean CBFV values were determined. The beat-to-beat
mean values of ABP and CBFV were exported to Matlab and resampled at 5 Hz
using spline-interpolation 24. From the ABP the mean value was subtracted and
the CBFVs were normalized with respect to the mean. This resulted in zero-
mean signals suited for spectral analysis to estimate the transfer function. The
transfer function was calculated by
S ( f )
H ( f ) ? xy (1)
S xx ( f )
whereSxx(f) is the auto spectrum of changes in arterial blood pressure and Sxy(f) is
the cross-spectrum between the ABP- and CBFV-signals. The transfer function
magnitude|H(f)| and phase spectrum (f) were derived from the real partHR(f)
and imaginary partHI(f) of the complex transfer function as
| H ( f ) |? H R ( f )_ 2 - H I ( f )_2 (2)
H ( f )
H( f ) ? arctan I (3)
H R ( f )
The squared coherence function 2(f) was estimated by
| S ( f ) |2
i 2 ( f ) ? xy (4)
S xx ( f )S yy ( f )
whereSyy(f) is the auto spectrum of changes in cerebral blood flow velocity. The
squared coherence reflects the strength of the linear relationship between ABP
and CBFV for each frequency on a scale from 0 to 1.
For the estimation of the transfer function ensemble averaging was used with
half overlapping windows with a length of 256 data points. With a 5 Hz sampling
rate this resulted in 51.2 seconds long data epochs. The smoothness priors ap-
proach 33 was used to remove the very low frequency trend. A hanning-
window was applied on each data epoch. The spectrum was smoothed with a
triangular window of 3 points wide corresponding to a bandwidth of 0.04 Hz.
47

proefschrift gommer inhoud met kaft.indd

To see the actual publication please follow the link above