Since its invention more than 20 years ago, functional Magnetic Resonance Imaging (fMRI) has played a central role in cognitive neuroscience. In recent years, ultra-high field MRI scanners (>= 7 Tesla) have become available, producing data at previously unobtainable quality.
At present however, the full complexity of fMRI data is poorly understood. Our own previous research, as well as that of several other research groups around the world, has shown that only a small percentage of the variance of fMRI time courses is explained by standard analysis techniques. Furthermore, task-related brain activation appears to be much more global and distributed than previously believed. Ultra-high field data show much more detail, making it even more difficult to understand and interpret the full information content.
The aim of this project is to develop new mathematical methods for the analysis of both standard fMRI data (3 Tesla) as well as ultra-high field fMRI data (>/=7 Tesla). We aim to harvest as much of the information content of fMRI as possible in order to advance our understanding of human brain function. Specifically, our work has focused on the development of novel techniques for statistical inference and for the detection of functional brain networks. Furthermore, we are currently developing machine learning techniques using deep neural nets, both considering the preprocessing and analysis of fMRI data.
The formation of transient networks in response to external stimuli or as a reflection of internal cognitive processes is a hallmark of human brain function. However, its identification in fMRI data of the human brain is notoriously difficult. Networks consist of nodes and edges connecting them, where nodes correspond to voxels in fMRI data, and the weight of an edge is determined via task-related changes in dynamic synchronisation between their respective times series. Based on these definitions, we developed a new data analysis algorithm that identifies edges in a brain network that differentially respond in unison to a task onset and that occur in dense packs with similar characteristics. Hence, we call this approach ``Task-related Edge Density'' (TED). TED proved to be a very strong marker for dynamic network formation that easily lends itself to statistical analysis using large scale statistical inference. A major advantage of TED compared to other methods is that it does not depend on any specific hemodynamic response model, and it also does not require a pre-segmentation of the data for dimensionality reduction, as it can handle large networks consisting of tens of thousands of voxels. Thus, this method can be applied to ultrahigh resolution fMRI data acquired at 9.4 Tesla without loss of spatial acuity.
Statistical inference in fMRI data analysis remains a challenging problem. Some currently used methods for statistical inference require spatial smoothing during preprocessing, which makes them unsuitable for use on ultrahigh resolution data because smoothing diminishes spatial accuracy. Furthermore, several of the most widely used statistical inference procedures were recently reported to produce inflated false positives rates. To address these issues, we have recently developed a new general framework for statistical inference. It is called "LISA" because it is inspired by hot spot analysis of geographical information systems, where hot spots are identified using so-called Local Indicators of Spatial Association (LISA, Anselin 1995). It does not require spatial smoothing during preprocessing, so that the spatial acuity is preserved. LISA is applicable in a wide range of application domains including ultrahigh resolution single subject data analysis, group data analysis, MVPA and many more. It boosts statistical power via edge-preserving filtering and controls the false discovery rate using random permutations. It shows higher sensitivity and lower false positive rates than commonly used methods of statistical inference
As discussed above, it is still a matter of debate how much information about the cognitive state is present in data acquired by means of functional magnetic resonance imaging. Machine learning methods offer an interesting perspective on this issue: for instance, pattern-based approaches can learn to capture information from small-grained patterns of brain activity, which would be invisible given traditional analysis approaches where only large-scale increases in the brain activity levels are considered. Recently, deep neural networks have become more widespread and have shown very promising results, particularly in computer vision. One of the main advantages of deep learning approaches is their ability to automatically learn and derive features during training. Thus, it is not necessary to manually set up features for learning. This in turn allows us to extract knowledge about which sources of information are present. The critical prerequisite, however, for such learning-based methods is the availability of large data sets. Thanks to recent efforts in big data initiatives such as the human connectome project and the availability of large-scale functional data at 9.4T, we are able to take a fresh look on functional imaging data. We are currently using deep learning methods for several projects, mainly for the preprocessing of fMRI. We have developed a learning-based approach for dealing with head motion correction, yielding higher precision than traditional methods. This is particularly needed for ultra-high resolution data as acquired by our 9.4T scanner.
Compartmentation and Connectivity of the Thalamus: fMRI "Resting State" and DWI Examinations
The human thalamus in terms of its internal parcellation, its connectivity patterns, the functioning of its circuitry, and its relationship to the cerebral cortex, still partly constitutes a terra incognita. Almost all information processing in cortex strongly depends on the thalamic interactions. Therefore, the knowledge of thalamic connections and interconnections is important to understand cortical functions. Although changes in the thalamus play a prominent role in the functionally defined pathophysiology of psychiatric and neurodegenerative diseases, its internal structures are largely identified and delineated based on structural cytoarchitectonic postmortem atlases, which are used for localization of neuro- or radiosurgical interventions. As such, there is an urgent need to understand the functional and structural subdivisions of the thalamus in vivo and provide a valid map for scientific and clinical studies. The objective of our study is to examine the subdivisions of the thalamus and determine its cortical connectivity pattern by the use of resting state fMRI (rsfMRI) and diffusion weighted imaging (DWI). First, we want to provide an in vivo segmentation of the human thalamus, which can be used for the analysis and interpretation of imaging findings in the clinical neurosciences. Secondly, this compartmentation will be analyzed in respect to its connectivity with cortical structures by means of correlation analysis and fiber tracking (FT). In addition, specific hemispheric and gender differences will be characterized in terms of the ensuing segmentations. Finally, in a third step, the functional and structural segmentation of the thalamus will be compared and related to the existing anatomical atlases. Overall, we hope that this project will enhance our understanding of the role of the thalamus in concert with the sensory and cognitive functions of the brain.
Lipsia software package
Lipsia is a software package for the analysis of fMRI data. It was originally developed at the MPI for Human Cognitive and Brain Science in Leipzig as a general purpose tool covering the entire range of fMRI data processing. In the last few years, the focus in Lipsia has shifted towards providing highly innovative new techniques, such as TED (Task-related edge density), ECM (Eigenvector centrality mapping) and statistical inference (LISA). The software is available at Opens external link in new windowhttps://github.com/lipsia-fmri/lipsia