Fusion of Electromagnetic Brain Imaging and fMRI

描述（由申请人提供）：多模态非侵入性功能脑成像在提高我们对人类行为的神经相关性的理解方面产生了巨大的影响，现在是系统和认知神经科学家不可或缺的工具。我们建议开发最先进的多模态功能成像融合算法，以实现大脑动态活动的精确可视化和高空间和时间分辨率。我们建议开发一种算法，将功能性磁共振成像（fMRI）的互补高空间分辨率和脑磁图（MEG）和脑电图（EEG）数据的高时间分辨率相结合，以实现大脑活动的高保真重建。近年来，我们的研究小组开发了一套新颖而强大的MEG/EEG成像算法，优于现有的基准算法，并将这些结果与皮质电图（ECOG）进行了比较。具体来说，我们的算法可以解决许多脑源，包括远离传感器的源，在使用快速和鲁棒概率推理技术的情况下，存在来自不相关脑源的大干扰。在这里，我们建议将M/EEG逆算法的成功扩展到多模态成像数据融合领域。我们的总体目标是利用最先进的机器学习算法，从嘈杂的MEG/EEG和fMRI数据中，最终产生亚毫米和亚毫秒分辨率的鲁棒性、高保真度的事件相关大脑激活视频。具体来说，我们建议将我们最近开发的一个强大的新算法（称为Champagne）扩展为两种新的融合算法，以不同的方式结合fMRI、MEG和EEG数据。这两种算法的性能将首先在模拟中进行严格评估，包括与现有基准融合算法的性能比较。然后，算法将在来自健康对照的四个fMRI-MEG+EEG数据集上进行一致性测试，这些数据集来自相同的范式（听觉、运动、图片命名和动词生成）和两个fMRI-EEG数据集（面部和运动感知）。还将对从癫痫患者获得的fMRI-MEG/EEG数据集进行额外的验证研究，并与皮质电图（ECoG）进行比较。在成功的测试和评估之后，本拨款提案中开发的所有算法以及示例验证数据集将使用我们开发的开源软件工具箱NUTMEG （nutme.berkeley.edu）进行分发。

DESCRIPTION (provided by applicant): Multimodal non-invasive functional brain imaging has made a tremendous impact in improving our understanding of the neural correlates of human behavior, and is now an indispensable tool for systems and cognitive neuroscientists. We propose to develop state-of-the-art multimodal functional imaging fusion algorithms for accurate visualization of the brain's dynamic activity and high spatial and temporal resolution. We propose to develop algorithms that combine complementary high spatial resolution of functional MRI (fMRI) and high-temporal resolution of magnetoencephalography (MEG) and electroencephalography (EEG) data for high-fidelity reconstruction of brain activity. In recent years, our research group has developed a suite of novel and powerful algorithms for MEG/EEG imaging superior to existing benchmark algorithms, and we have compared these results with electrocorticography (ECOG). Specifically, our algorithms can solve for many brain sources, including sources located far from the sensors, in the presence of large interference from unrelated brain sources using fast and robust probabilistic inference techniques. Here, we propose to extend this success in M/EEG inverse algorithms into the domain of multimodal imaging data fusion. Our overall goal here is to ultimately produce robust, high fidelity videos of event-related brain activation at a sub-millimeter and sub-millisecond resolution from noisy MEG/EEG and fMRI data using state-of-the-art machine learning algorithms. Specifically, we propose to extend a powerful new algorithm that we have recently developed, called Champagne, into two new fusion algorithms that combine fMRI, MEG and EEG data in different ways. Performance of both algorithms will first be rigorously evaluated in simulations, including performance comparisons with existing benchmark fusion algorithms. Algorithms will then tested for consistency on four fMRI-MEG+EEG datasets from healthy controls obtained for identical paradigms (auditory, motor, picture naming and verb-generation) and two fMRI-EEG datasets (face and motion perception). Additional validation studies will also be performed on fMRI-MEG/EEG datasets obtained from epilepsy patients and compared to electrocorticography (ECoG). Following successful testing and evaluation, all algorithms developed in this grant proposal, as well as example validation datasets, will be distributed using NUTMEG (nutmeg.berkeley.edu), an open-source software toolbox that we have developed.