Brain Image Registration Using Neuro Fuzzy Entropy

Abstract

In this research work, brain image slices are considered for newlineregistration task. Hilbert-Huang Transform (HHT), Fuzzy logic, Cerebellar newlineModel Articulation Controller (CMAC) neural network are used for newlineregistration of the brain image slices. As a convention, floating image has newlinemisalignment. The floating image has to be transformed and aligned with newlinethe target image. newlineTwo approaches can be used to create floating image. The first newlineapproach is to use the actual image obtained in scanning and it is newlineconsidered as target image. The second approach is to introduce newlinemisalignment in an existing image slice to obtain a floating image. newlineThis thesis presents ANN and Fuzzy logic approach as an alternative newlinefor the conventional (existing) algorithms for registering functional newlinemagnetic resonance (fMRI) image slices. The application area considered newlineis, registration of image slices of human brain acquired using magnetic newlineresonance imaging scanner. newlineIn this work, a systematic approach has been developed for newlineregistration of the fMRI image slices of one session and in particular fMRI newlineimage slices of different sessions (rest-task-rest-task). newlineTwo types of features are used for registration. newline1. Features based on the pixel coordinates newline2. Features based on the Hilbert-Hang Transform(HHT) newlineoutput newlinea) Features based on the pixel coordinates are as follows: newline1. Vertical shift {Upward (1) or Downward (2)}. newline2. Horizontal shift {Left (1) or Right (2)}. newline3. Angle rotated {Direction CW (2) or CCW (1)}. newlineThese features are used as training patterns for ANN algorithms. newlineb) Features based on HHT newlineHuang developed an empirical method of extracting various signal newlinecomponents present in a single signal. He combined the concept newlinedeveloped by Hilbert to extract instantaneous frequency and newline7 newlineinstantaneous amplitudes. The combined method of Huang and Hilbert is newlinecalled Hilbert Huang transform. The features obtained from HHT are as newlinefollows. newline1. Mean of Instantaneous Amplitude and Instantaneous newlinefrequency. newline2. Maximum of Instantaneous Amplitude and Instantaneous newlinefrequency. newline3. Minimum of Instantaneous Amplitude and Instantaneous newlinefrequency. newline4. Norm of Instantaneous Amplitude and Instantaneous newlinefrequency. newline5. Standard deviation of Instantaneous Amplitude and newlineInstantaneous frequency. newline6. Coarse to fine and newline7. Fine to coarse energy value. newlineThe features extracted from the pixel coordinates / features newlineextracted from the HHT are presented to CMAC and Fuzzy logic algorithms newlinefor an improved registration of the images. newlinePurpose of using CMAC and Fuzzy logic in the research work is newlinebecause, existing methods are based on statistical parameters to register newlineimages to the expected accuracy. The purpose of using CMACand Fuzzy newlinelogic for image registration is due to the following reasons: newline1. The working concepts of CMAC are based on statistics like using newlinelinear summation between layers to propagate information from input to newlineoutput layers, newline2. CMAC uses quantization and objective function for finding optimal newlineweights between layers for mapping inputs (floating image) to newlineoutputs (Target image). newlineBecause of the working properties of CMAC are based on statistical newlineconcepts, CMAC assures correct registration. newline8 newlineCMAC is the mathematical representation of the functioning of newlineneural connections in the human brain. The mathematical representation newlinevaries depending upon the application. newlineFuzzy logic has been used due to its property like membership newlinefunction that helps in improved image registration. newlineJustification of methods implemented in this thesis with respect to newlineproblem of rigid registration is based on the following: newlinea) Rigid registration involves change in position / orientation image 1 newlinewith image 2 for correct alignment. The shape and size of the newlineobjects of the images are not changed. In this work, no scaling and newlineshearing is involved. Only translation (x, y), horizontal and vertical newlinedirection and rotation in one plane are used. newlineThe thesis mainly contributes the following methods for improved newlineregistration: newline1. Applying HHT for extracting features from the image 1 and image 2. newline2. Applying CMAC neural network for mapping the features of image 1 to newlinefeatures of image 2 and resulting in improved image registration. newline3. Applying Fuzzy logic for mapping the features of image 1 to features of newlineimage 2 and resulting in improved image registration. newline