Research On Limited Angular Reconstruction Algorithms For Linear Scan CT

Traditional computed tomography(CT) is not appropriate for applications which need rotations. Linear scanning computed tomography(LCT) in a straight line trajectory shows promising applications of medical imaging, industrial and security inspection because of its high scanning efficiency. However, limited by the ray-beam flare angle and the detector size, the available scanning angle is within a very limited range, resulting the image reconstruction a limited angle problem. Therefore, research on LC T image reconstruction algorithm is one of the hottest spot and of significant value in both theoretical and practical aspects.Considering the shortages on accuracy and efficiency of the art methods, this dissertation focuses on the reconstruction for LC T in both spatial and frequency domain, together with the parallel acceleration of reconstruction algorithms. The main research is included as follows:1. A spatial iterative reconstruction algorithm named EGTVM(edge guided total variation minimization) is proposed, which is based on iterative edge detection and weighted total- variation(TV) regularization. EGTVM is based on the framework of iterative support detection(ISD) theory, utilizing the iterative edge detection and weighted TV minimization in an alternating scheme. The alternating direction method(ADM) is applied to solve the weighted TV minimization. Experimental results indicate that EGTVM converges faster and reconstructs more accurate than the state of the art algorithms.2. A Fourier-based iterative reconstruction algorithm based on the framework of ADM named PPF-TVM(pseudo polar Fourier-based total variation minimization) is proposed, utilizing pseudo polar fast Fourier transform and TV minimization. A Fourier-based TV minimization optimization model is established and the ADM scheme is used to solve the model. PPF-TVM adopts the pseudo polar fast Fourier transform(PPFFT) to avoid the interpolation in Fourier domain, making the reconstruction both efficient and robust. Experimental results show that the PPF-TVM algorithm has outstanding advantages in accuracy over traditional Fourier-based algorithms while with high computation efficiency.3. An acceleration method for EGTVM is proposed, based on fast forward- and backward-projection and graphic processing unit(GPU). Considering that the spatial iterative algorithm has a tremendous computation cost, a fast forward- and backward-projection technique for LCT reconstruction, with its efficient implementation and optimization on GPU, is developed based on the recent research on x-ray transform. Experimental results show that EGTVM gets an excel ent acceleration ratio of about 100 without sacrificing the reconstruction accuracy.