Understanding and reducing artefacts in simultaneously acquired EEG and fMRI data

在脑电图（EEG）中，通过测量神经元在头皮表面产生的微弱电压来监测大脑活动。脑电图测量可以在毫秒的时间尺度上进行，因此对于理解大脑反应的时间是有用的，但是要计算出电压是从大脑的哪里产生的并不容易。功能性磁共振成像（fMRI）可以高精度地识别大脑活动的部位，但fMRI并不能提供有关大脑反应时间的信息，因为它是基于血流相对缓慢变化的影响。EEG和fMRI的互补属性意味着它们在同步EEG-fMRI中的组合可能非常有用，但将这两种技术结合起来在技术上具有挑战性，因为由于大脑活动产生的电压远远小于MR扫描仪内发生的大时变磁场产生的伪影。伪影的最大来源是形成MR图像所需的磁场梯度的快速切换。由此产生的梯度伪影（GA）可能比大脑信号大10，000倍。由于梯度波形是周期性的，因此可以形成平均伪影模板，可以从每个伪影发生中减去该平均伪影模板以清理EEG记录。然而，如果受试者在扫描期间移动并且需要具有非常大的动态范围的EEG系统来记录GA，则这种平均伪影减法（AAS）失败。第二个伪影通常比大脑信号大10-100倍，与心动周期有关。已经提出了这种脉冲伪影（PA）的几种可能的来源，包括由心脏脉动驱动的头部旋转和头皮扩张，以及由于磁场中血液的脉动流而引起的霍尔电压。PA的周期性性质意味着它也可以使用AAS进行校正，但是PA通常在心跳之间显著变化，使得难以使用AAS完全消除这种伪影。 在同步fMRI过程中EEG记录中残留的GA和PA的存在限制了组合EEG-fMRI的应用，特别是在研究产生弱或高频信号的大脑活动时。因此，这里提出的工作的目的是开发设备和技术，这将提高质量的EEG数据采集与并发功能磁共振成像，从而允许的全部潜力相结合的EEG功能磁共振成像得以实现。聚焦于GA，我们将识别和减少EEG系统的不同组件对伪影的贡献，然后识别受试者头部在扫描仪中的方向和位置，将GA的影响降低到最低水平。我们将开发和测试新的校正方法，以抵消在扫描过程中受试者移动时发生的GA变化的影响。计算机建模和实验将用于优化连接到EEG电极的导线的布局，以减少GA。增加一个参考层，经历类似的伪影电压在头皮上产生的好处也将被调查。在PA上，我们将识别不同伪影来源的相对贡献，然后使用此信息优化EEG导线的布局，并测试使用参考层和来自连接到头部的运动传感器的信息在减少PA中的益处。关于GA和PA的工作结果将被应用于改进消除后处理中两种伪影的方法，并将被汇集在一起以确定最佳实验设置，该设置将在与神经科学家一起进行的实验中进行测试。 这些发展将为许多研究人员提供直接的好处，这些研究人员使用组合EEG-fMRI来研究正常大脑和神经系统疾病（包括癫痫和精神分裂症）的脑功能变化。

In electroencephalography (EEG), brain activity is monitored by measuring weak voltages produced at the surface of the scalp by neurons. EEG measurements can be made on a millisecond timescale and so are useful for understanding the timing of brain responses, but it is not easy to work out from where in the brain the voltages arise. Functional magnetic resonance imaging (fMRI) allows the site of brain activity to be identified with high accuracy, but fMRI does not provide much information about the timing of brain responses because it is based on effects of relatively slow changes in blood flow. The complementary attributes of EEG and fMRI mean that their combination in simultaneous EEG-fMRI is potentially very useful, but combining the two techniques is technically challenging because the voltages due to brain activity are much smaller than the artefacts produced by the large time-varying magnetic fields that occur inside an MR scanner. The largest source of artefact is the rapid switching of magnetic field gradients, needed to form MR images. The resulting gradient artefact (GA) can be 10,000 times larger than the brain signals. Since the gradient waveforms are periodic, it is possible to form an average artefact template that can be subtracted from each artefact occurrence to clean up the EEG recording. However, this average artefact subtraction (AAS) fails if the subject moves during the scanning and an EEG system with a very large dynamic range is needed to record the GA. The second artefact, which is typically 10-100 times larger than brain signals, is linked to the cardiac cycle. Several possible sources of this pulse artefact (PA) have been proposed, including head rotation and scalp expansion driven by cardiac pulsation, and Hall voltages due to pulsatile flow of blood in the magnetic field. The periodic nature of the PA means that it can also be corrected using AAS, but the PA often varies significantly across heartbeats making it difficult to completely eliminate this artefact using AAS. The presence of residual GA and PA in EEG recordings made during simultaneous fMRI limits the application of combined EEG-fMRI, particularly in studying brain activity that produces weak or high frequency signals. The aim of the work proposed here is therefore to develop equipment and techniques which will improve the quality of EEG data acquired with concurrent fMRI, thus allowing the full potential of combined EEG-fMRI to be realised. Focusing on the GA, we will identify and reduce the contributions of different components of the EEG system to the artefact and then identify the orientation and position of the subject's head in the scanner that reduces the effect of the GA to its lowest level. We will develop and test new correction methods for counteracting the effects of changes in the GA that happen when the subject moves during a scan. Computer modelling and experiments will be used to optimise the lay-out of the wires linking to the EEG electrodes so as to reduce the GA. The benefits of adding a reference layer which experiences similar artefact voltages to those produced at the scalp will also be investigated. On the PA, we will identify the relative contributions of the different sources of the artefact and then use this information to optimise the lay-out of the EEG wires and to test the benefits of using a reference layer and information from movement sensors attached to the head in reducing the PA. The findings of the work on the GA and PA will be applied to improving methods for eliminating both artefacts in post-processing and will be brought together to identify an optimal experimental set-up which will be tested in experiments carried out in conjunction with neuroscientists. The proposed developments will provide immediate benefit to the many researchers who use combined EEG-fMRI in studying the normal brain and changes in brain function in neurological disorders, including epilepsy and schizophrenia.