Super-resolution Reconstruction In Medical Image

Most medical imaging devices acquire human body image data by radiant energy. To maximize image resolution, super resolution technique combines multiple low-dosage images, each containing spatial shifts to reconstruct a super-resolution image. These shifts are the result of patient movement, intentional dithering of the detector, vibration in the imaging system, and small movement of the imaging gantry. Improving the resolution of medical imaging devices is of great importance to the medical community since early detection and intervention often results in optimal treatment and recovery.Firstly, the mathematical model of SR is introduced. The point spread function(PSF) of X-ray computed tomography(CT) is discussed, and is satisfying the iterative back projection(IBP) approach requirement for convergence. The practicability for iterative back projection approach implement in CT image is discussed, and the mathematical model of SR for CT is introduced.We adopt the phase correlation algorithm base on the property of Fourier transform to solve the image registration problem of CT image. How phase correlation estimate the relative offsets between similar images is discussed. It is possible to represent rotation as shifts by converting from rectangular coordinates to log-polar coordinates, so that phase correlation estimate the rotation between similar images. Experimental results demonstrate that phase correlation is effective in image registration, and it has higher registration precision and strong robustness, and it meets the super resolution reconstruction requirements.Two improved interpolation algorithm is proposed. One is nearest neighbor interpolation combined with sub-pixel shifts. Another is bilinear interpolation combined with sub-pixel shifts. Both can be used in iterative back projection. An improved iterative back projection algorithm is proposed. Combined with sub-pixel shifts, the degraded process and error data back projection in iterative process is improved. More information is added to reconstruct the high solution image, so that reconstruction image contains more texture structures and details.Experimental results on super-resolution in CT demonstrate the practicability for super-resolution technique apply in medical imaging. The algorithm in this thesis can reconstruct the high resolution image with texture structures and details, avoided over-smoothing made by the iterative process.