A generative approach to human brain mapping

点击翻译按钮获取中文摘要

Functional magnetic resonance imaging (fMRI) provides an unparalleled opportunity to observe the activity in the human brain during any mental activity. Scientists are working on to use this technique to understand the neuronal processes that give rise to intelligent behavior, as well as understanding and predicting neuro-psychiatric diseases such as Schizophrenia or Alzheimer's. Despite its promise as diagnostic tool, however, fMRI is currently not broadly used in clinical practice. One important road block is that even normal, healthy individual brains have quite different spatial layouts and that we need a lot of data to reliably characterize it. Just like trees in a forest show a wide array of spatial arrangements of branches and leaves, human brains show a wide variety of spatial arrangement of different functional regions. This variability makes it very difficult to draw conclusions about the brains of specific groups of subjects, as we lose a lot of information when averaging data across brains. The large variability also makes it difficult to detect unusual brain organization. The long-term goal of my research program is to build a model of human brain organisation that captures the variability observed in brain maps measured with fMRI. The model has two parts. The spatial arrangement model captures the average layout of different brain regions in the human brain, as well as the variability. The second part models how this arrangement influences the observed data, be it from resting-state or task-based paradigms. This has the big advantage that the model can integrate information over many openly available functional imaging studies, and thus can be trained on data from many thousands of participants with different genders, ages, and ethnicities. The first part of the proposal will address the algorithmic and computational challenges of building such a model. In the second part, we will validate the model by testing its ability to predict the location of functional regions in individuals. Once completed, the model allows us to obtain a high-quality map of a brain from an individual, even though we may have 10-30min of data. Currently, longer scan times are necessary to obtain a reliable functional characterization of an individual brain. This model will be useful for 3 important applications: First, it will allow us to draw better conclusions from group studies, as we will be able to account for the individual variability within those groups. Second, we will be able to obtain a highly detailed map of individual brains even if we have only restricted data available. This makes individual functional mapping feasible, which then can be used in pre-surgical planning for brain tissue removal or implantation of brain stimulators. Finally, the model will help us to spot abnormal brain organisation more easily, to become better at predicting neuropsychiatric diseases, and to understand the underlying fundamental brain processes.