Research On Pattern Analysis Method Of FMRI And OI Dataset

Three kinds of pattern analysis method are proposed for functional magnetic resonance image(fMRI) and optical imging(OI) dataset in this paper. The first one is the local temporal independent component analysis technique; the second kind of technique is structured component analysis; the last kind of technique is rapid spectral functional optical image procedure.In fMRI and OI data, the spatial dimension is usually far higher than temporal dimension, and this will bring the illness into covariance matrix utilized by spatial PCA in dimension reduction. The illness of the covariance matrix will make the tICA either unfeasible or unstable. In this paper, two local temporal independent component analysis techniques are proposed which can reduce the spatial dimension efficiently and eliminate the illness of the covariance matrix. In techniques, spatial and temporal characteristics of stimuli-induced signal dynamic responses can be investigated simultaneously. For fMRI data, firstly the multitaper spectral estimation is utilized to estimate the spectrum of each voxel; the significance of the line frequency components at the interested frequency is tested to detect the task-related cortex areas; the temporal independent component analysis (tICA) is then applied to the activated voxels to obtain stimuli-induced signal dynamic responses. The advantage of this procedure is that little assumptions are needed for the cerebral hemodynamics and spatial distribution of task-related areas. Problems which were often met by tICA analysis of fMRI data, such as the lack of stability, reliability and robustness, are overcome by the suggested method. For OI data, the continuous wavelet transform (CWT) is exploited to detect the activated voxels of cortex and exploits temporal independent component analysis (tICA) to extract the underlying independent sources whose number is determined by Bayesian information criterion. The neural response signals and the intrinsic fluctuation signals are picked out from the independent sources by investigating their temporal architecture. The stimuli-induced signal dynamic responses together with the pulse-induced and the 0.1Hz fluctuation signals are extracted from data successfully. The procedure proposed is a valuable technique for researchers to investigate the temporal and spatial architectures of cortical functional mapping.In structured component analysis technique, the noise is classified into structured and unstructured one by its different characteristics of temporal autocorrelation. For fMRI data, the canonical correlation analysis (CCA) is exploited to extract the underlying independent sources from which the structured ones are recognised by the white noise (WN) criterion. The signal dynamic responses are discerned from the structured sources by surrogate test based on the reduced autoregression model (ST-RARM). The low order autocorrelation of the unstructured noise is eliminated by randomization technique and this will lead to a significant reduction of unstructured noise'spectral power in low frequency. The signal dynamic responses and the randomly permuted unstructured noise are used to reconstruct the dataset which will meet the assumption of the white background noise well. Twenty sets of true fMRI data are processed. Statistic values of the task-related voxels increase significantly and some task-related areas which can't be detected from the original dataset are discerned. For OI data, the multi-lags canonical correlation analysis (MLCCA) put forward in this paper is exploited to extract the underlying structured sources. The signal dynamic responses are discerned from the structured sources by surrogate test based on the reduced autoregression model (ST-RARM). The signal dynamic responses and the unstructured noise are used to reconstruct the dataset which will meet the assumption of the white background noise well in most statistic and inference, and the phase differences of voxels'signal dynamic responses are hold. Monte-Carlo simulation demonstrates the success of our procedure in OI noise reduction. Together with a reduction of recognition false rates, there is also a significant improvement of recognition accuracy rates. Five sets of true OI data are processed. Speckle noise in functional mapping is weakened effectively and the task-related voxels are more clustered. The pseudo-mapping introduced by vessel net is removed by considering the phase difference of voxels.In rapid spectral optical functional mapping technique, Continuous-periodic stimulation is applied to the animals. The period of the stimuli is much shorter than the duration of the hemodynamic course. The signal dynamic responses and its related metabolically induced changes will enter a stable state in about half minute after the onset of the stimuli. The map signals vibrate at the stimulation frequency round some lever in state space and this vibration can be detected by the spectral analysis techniques. Hind-Paw (HP) area of SD mouse with sciatic nerve stimulation is used as the neural model. The red light (605±10nm) is exploited as illumination. The response power maps together with the phase distribution are given by F-map based on multitaper spectral estimate and Fourier spectral estimate. The functional map is obtained by synthesizing the information of response power maps and phase distribution. The vascular overspill phenomenon is weakened greatly resulting in pseudo-maps reduction. The phase distribution indicates that response of contralateral HP area is temporally prior to that of ipsilateral HP area. The arterial-venous transit and the local blood circulation are also discussed in the paper. It is found in our experiment that when relatively high frequency stimuli are applied, the vessel net will resonate with the stimuli while the cortex will not.