| This study introduces new signal processing methods for extracting meaningful information from brain signals (functional magnetic resonance imaging and single unit recording) and proposes a content-based retrieval system for functional brain data. First, a new method that combines maximal overlapped discrete wavelet transforms (MODWT) and dynamic time warping (DTW) is presented as a solution for dynamically detecting the hemodynamic response from fMRI data. Second, a new method for neuron spike sorting is presented that uses the maximal overlap discrete wavelet transform and rotated principal component analysis. Third, a procedure to characterize firing patterns of neuron spikes from the human brain, in both the temporal domain and the frequency domain, is presented. The combination of multitaper spectral estimation and a polynomial curve-fitting method is employed to transform the firing patterns to the frequency domain. To generate temporal shapes, eight local maxima are smoothly connected by a cubic spline interpolation. A rotated principal component analysis is used to extract common firing patterns as templates from a training set of 4100 neuron spike signals. Dynamic time warping is then used to assign each neuron firing to the closest template without shift error. These techniques are utilized in the development of a content-based retrieval system for human brain data. |
