Research On Blind Equalization Algorithm In Medical CT Image Blind Restoration

Medical CT image is one of main basis for the diagnosis and treatment ofdisease, and its quality will directly affects the accuracy of diagnosis. However, in theprocess of medical CT imaging, due to the impact of the point spread function, imagewill emerge from degradation, and affect diagnosis effect. And the exact cause ofdegradation may be unknown. Blind image restoration algorithm is that the unknownimage, blur and all model parameters, including the noise variances, are estimatedsolely from the observations without prior knowledge or user intervention. It waswidely utilized in the field of astronomical imaging, medical diagnostics, military andpublic security. Preliminary results of blind equalization algorithm were first found inthe signal case then extended to the image. Reducing the affect of point spreadfunction is equivalent to eliminating the inter-symbol interference. It has become ahot research topic in the field of communication signal processing and imageanalysis.The major contribution of this paper is summarized as follow:（1） A medical CT image blind equalization algorithm based on dimensionreduction was proposed in this paper. We defined the cost function applied to medicalCT image and demonstrate the performance of convergence. At the same time, thestrategy of variable step size was utilized to solve the contradiction betweenconvergence speed and accuracy. Two medical CT image blind equalizationalgorithms based on reduction dimension were proposed. Simulation results show thatthe proposed algorithms have faster convergence rate and smaller steady state error.In order to speed up the convergence of constant modulus algorithm and improve theperformance of algorithm, the second rank Hessian information of cost function wasutilized in the process of weight update. Computer simulations demonstrate the newalgorithm improves the convergence of the algorithm performance and peak signalnoise ratio.（2） A medical CT image constant module blind equalization algorithm basedon row-column transform was proposed. The affect of point spread function was decomposed to vertical and horizontal direction. The characteristics that I and Qdirection does not affect each other in the complex blind equalization algorithm, wasutilized to eliminate the affect of point spread function. Iteration formula was derivedand the static and dynamical convergence performance was analyzed. It is shown in aseries of computer-simulated experiments that the proposed method outperforms anumber of existing alternatives in terms of peak signal to noise ratio and recoveryeffects. By use of frequency domain transform, the frequency domain minimum errorprobability medical CT image blind equalization algorithm was proposed, theselection principle of the step size was analyzed, and computer simulation verify thevalidity of the algorithm.（3） Both medical CT image neural network blind equalization algorithmsbased on Zigzag coding and double Zigzag coding were proposed. Three layer neuralnetwork structures were adopted. We designed the transfer function of neural network,derived iteration formula and analyzed convergence performance. Computersimulation experiments show that the proposed algorithm reduces mean square errorand improves restoration effect, peak signal to noise ratio and improving signal tonoise ratio.