Coll-Font J, Erem B, Štóvíček P, Brooks D H
B-spiral group, Northeastern University, Boston (MA), USA.
Computational Radiology Lab, Boston Children's Hospital, Boston (MA), USA.
Proc IEEE Int Symp Biomed Imaging. 2015 Apr 16;2015:1053-1056. doi: 10.1109/ISBI.2015.7164052.
Inverse methods for localization and characterization of cardiac and brain sources from ECG and EEG signals are notoriously ill-conditioned and thus sensitive to SNR in the measurements. Multiple recordings of the same underlying phenomenon are often available, but are contaminated by unmodeled correlated noise such as heart motion from respiration or superposition of atrial activation or on-going EEG in the case of inter-ictal spikes or evoked response in EEG. We address here the open question of how best to incorporate these multiple recordings, comparing standard ensemble averaging, a multichannel non-linear spline-based average designed to be less sensitive to timing variations from motion or modulation, and a probalistic inverse incorporating a data-driven model of the noise correlation and using all recordings jointly. Results are tested on localizations of clincally recorded 120 lead ECGs during ventricular pacing.
从心电图（ECG）和脑电图（EEG）信号中对心脏和脑源进行定位与特征描述的逆方法，其病态性众所周知，因此对测量中的信噪比很敏感。通常可以获得同一潜在现象的多次记录，但这些记录会受到未建模的相关噪声污染，例如呼吸引起的心脏运动、心房激活的叠加，或者在发作间期尖峰情况下的持续脑电图，又或者脑电图中的诱发反应。在此，我们探讨如何最好地整合这些多次记录这一开放性问题，比较标准总体平均法、一种基于多通道非线性样条的平均法（该方法设计为对运动或调制引起的时间变化不太敏感）以及一种概率逆方法（该方法纳入了噪声相关性的数据驱动模型并联合使用所有记录）。结果在心室起搏期间临床记录的120导联心电图定位上进行了测试。
