Spectral Analysis And Pattern Recognition Of Brain Functional Signals In Optical Imaging

With optical imaging of intrinsic signals (OIS), the investigation content of thisdissertation contains two aspects:(1) we investigated the spectral feature distributionsof intrinsic low frequency oscillation signal, respiration signal and heartbeat signal inthe cerebral cortex and their application in cortical artery-vein separation.(2) By usingintraoperative OIS, we investigated the phenomenon of multisensory processing andintegration in the auditory cortex from patients. The experimental result provides a newevidence for the theory that multisensory processing and interaction occur at the earlystage.The cerebral active signals recorded by OIS consist of abundant intrinsic signalcomponents. Analyzing the power spectral distribution and spatiotemporal features ofthese signal components is very important for understanding the brain functionalactivity mechanism. Furthermore, determining the physiological sources of the intrinsicoptical signals is also one of the important research goals in brain functional opticalimaging. This dissertation investigated the spectral distribution features of physiologicalsignals from the optical image sequences in cerebral cortex, and found that the powerspectral distribution of0.1-Hz low frequency oscillation, respiration and heartbeatsignals could reflect specific vessel network structures, which presented valuableapplication in identifying arteries and veins. By utilizing these features and combiningwith image processing techniques as well as pattern recognition approaches, weproposed three cerebral artery-vein separation methods:(I) The power spectra of the intrinsic optical signals were firstly estimated by themultitaper method. Then a standard F-test was performed on each discrete frequencypoint to test the statistical significance at the given level. Four periodic physiologicaloscillations were examined: heartbeat frequency (HF), respiration frequency (RF), andtwo other eigenfrequencies termed F1and F2. In arterial regions, the spectral power ishigher in HF and F1, whereas in venous regions, the spectral power is higher in RF. F2exhibits higher spectral power in microvessels. The separation of arteries and veins wasimplemented with the fuzzy c-means clustering method and the region-growingapproach by utilizing the spectral amplitudes and power-ratio values of the foureigenfrequencies on the vasculature.(II) With546-nm and630-nm wavelengths illuminating the cerebral cortex, wefound that the0.1-Hz oscillation at546nm exhibits greater amplitude in arteries than inveins, whereas the0.1-Hz oscillation at630nm exhibits greater amplitude in veins thanin arteries. This spectral feature enables cortical arteries and veins to be segmentedindependently by combinging with multiscale matched filters of a modified dualGaussian model and single Gaussian model. Our method can separate most of the thin arteries and veins from each other, especially the thin arteries with low contrast in rawgray images. In vivo OIS experiments demonstrate the separation ability of the0.1-Hzbased segmentation method in cerebral cortex of eight rats. The validation resultsindicate that our0.1-Hz method is very effective in separating both large and thinarteries and veins regardless of vessel crossover or overlapping to great extent incomparison with previous methods.(III) We utilizied canonical correlation analysis (CCA) and independentcomponent analysis (ICA) to decompose the intrinsic optical signals in temporal andspatial domain. Three distinct signal components including low frequency oscillation,respiration and heartbeat were selected from the decomposed components as the featurevector. Low frequency oscillation and heartbeat sources reveal the arterial structurewhile respiration source reveals the venous structure. The signal components intemporal domain should be firstly constructed as correlation-coefficient map to revealthe specific vessel spatial structure, while the components in spatial domain could beused as feature vector directly. Based on the three feature maps, classification of vesseltypes is achieved by support vector machine (SVM) on segmented vessel network. Withhand-labeled arteries and veins as the reference standard, performance of this methodwas tested and compared with previously reported methods. The results showed thattemporal CCA and temporal ICA methods worked better on extracting respiration signalfor highlighting veins and inhibiting arteries than the spatial methods, as well as singlefrequency method, while the spatial CCA and spatial ICA methods worked better onextracting heartbeat signal for highlighting arteries and inhibiting veins than thetemporal methods.On another aspect, we detected the activation of the superior temporal gyrus underthe acoustic stimuli and sensory stimuli with intraoperative OIS. OIS can provide higherspatial and temporal resolution than fMRI, thus can help us deeply understand thedetailed hemodynamic response of the region of interest (ROI). The primary auditorycortices (Brodmann’s Area, BA41&42) were found to respond significantly to thesensory stimulation and the strongest activation area for sensory stimulation was locatedin BA42, suggesting that multisensory processing exists in the auditory cortex ofhuman beings. Our finding provides a new evidence for the theory that multisensoryprocessing and interaction occur at the early stage.