背景:血液动力学信号的同步采集对于其多模态分析至关重要,例如动脉血压(ABP)的动态脑自动调节(DCA)分析和经颅多普勒(TCD)衍生的脑血流速度(CBv)。几个技术问题可以,然而,导致不同信号之间的(变化的)时移。这些可能难以识别并且可能强烈影响多模态分析结果。
方法:我们开发了一种多步骤,多模态脉动血流动力学信号的基于互相关的时移检测和同步算法。我们已经从包含几个时移的组合的数据集中使用ABP和CBv测量开发了算法。我们在具有时移的外部数据集上验证了该算法。我们还对算法在人工添加时移的数据集上的性能进行了定量验证,由-0.2到0.2s/min的采样时钟差和-4到4s之间的突然时移组成。量化了叠加噪声和波形形态变化对时移估计的影响,并确定了它们对DCA指数的影响。
结果:人为添加的时移与估计的时移之间的瞬时中值绝对误差(MedAE)为12ms(中位数,IQR12-12,范围为11-14ms),用于-0.1和0.1s/min之间的漂移以及-4和4s之间的突然时移。对于高于0.1s/min的漂移,MedAE较高(中位数753,IQR19-766,范围13-772ms)。当包括确定性阈值时(峰值互相关>0.9),所有漂移-移位组合的MedAE降至12ms,具有较小的变异性(IQR12-13,范围8-22ms,p<0.001)。时移估计对噪声具有鲁棒性,因为对于方差等于信号方差的叠加白噪声,MedAE相似。时移校正后,DCA指数与原始指数相似,非时移信号。相移相差0.17°(中位数,IQR0.13-0.2°,范围0.0038-1.1°)和0.54°(中位数,IQR0.23-1.7°,范围0.0088-5.6°),适用于极低频和低频范围,分别。
结论:该算法允许视觉上可解释的检测和对脉动血液动力学信号(ABP和CBv)之间的时移的精确校正。
BACKGROUND: Synchronous acquisition of haemodynamic signals is crucial for their multimodal analysis, such as dynamic cerebral autoregulation (DCA) analysis of arterial blood pressure (ABP) and transcranial Doppler (TCD)-derived cerebral blood velocity (CBv). Several technical problems can, however, lead to (varying) time-shifts between the different signals. These can be difficult to recognise and can strongly influence the multimodal analysis results.
METHODS: We have developed a multistep, cross-correlation-based time-shift detection and synchronisation algorithm for multimodal pulsatile haemodynamic signals. We have developed the algorithm using ABP and CBv measurements from a dataset that contained combinations of several time-shifts. We validated the algorithm on an external dataset with time-shifts. We additionally quantitatively validated the algorithm\'s performance on a dataset with artificially added time-shifts, consisting of sample clock differences ranging from -0.2 to 0.2 s/min and sudden time-shifts between -4 and 4 s. The influence of superimposed noise and variation in waveform morphology on the time-shift estimation was quantified, and their influence on DCA-indices was determined.
RESULTS: The instantaneous median absolute error (MedAE) between the artificially added time-shifts and the estimated time-shifts was 12 ms (median, IQR 12-12, range 11-14 ms) for drifts between -0.1 and 0.1 s/min and sudden time-shifts between -4 and 4 s. For drifts above 0.1 s/min, MedAE was higher (median 753, IQR 19 - 766, range 13 - 772 ms). When a certainty threshold was included (peak cross-correlation > 0.9), MedAE for all drifts-shift combinations decreased to 12 ms, with smaller variability (IQR 12 - 13, range 8 - 22 ms, p < 0.001). The time-shift estimation is robust to noise, as the MedAE was similar for superimposed white noise with variance equal to the signal variance. After time-shift correction, DCA-indices were similar to the original, non-time-shifted signals. Phase shift differed by 0.17° (median, IQR 0.13-0.2°, range 0.0038-1.1°) and 0.54° (median, IQR 0.23-1.7°, range 0.0088-5.6°) for the very low frequency and low frequency ranges, respectively.
CONCLUSIONS: This algorithm allows visually interpretable detection and accurate correction of time-shifts between pulsatile haemodynamic signals (ABP and CBv).