Based Brain Function Resting State FMRI Research Data Connection

Resting state functional magnetic resonance imaging (rs-fMRI) reflects brain activity in the resting state, and it has been widely used to study human brain function. In this dissertation, we studied two major parts of the human brain function based on rs-fMRI data. First, we proposed a new method for constructing brain functional sub-networks based on sparse representation. Wavelet transform coherence (WTC) is defined as the measure of functional interaction pattern between different ROIs (region of interest), features of each ROI are extracted with sparse reperesentation afterwards, spectral clustering is excuted for heuristic clustering lastly. We performed this method with10healthy subjects’rs-fMRI data and8sub-networks were constructed. The result shows these8sub-networks concide with brain’s functional systems such as vision, sensory-motor et al. and are consistent with current neuroscience knowledge. Besides, two traditional methods were adopted to build brain sub-networks with the same data, by contrast, the method proposed in this dissertation shows a better result.Inferring characteristic functional connections of healthy people and mild cognitive impairment (MCI) patients using sparse representation is the second part of this dissertation. First, we calculate WTC of each pair of ROIs as the metric of functional connection, dictionary learning and sparse coding are performed for each subject afterwards, common functional connections are extracted for healthy and MCI subjects respectively at last. We studied63subjects of two different datasets to conclude a more common method. We got satisfactory results when using these characteristic functional connections for SVM classification. Finally, to further validate the effectiveness and robustness of the method, a series of artificial data were generated to run the same method. The result proves that our method is effective and robust.