Application Research On Compressed Sensing Based CT Iterative Image Reconstruction

CT is one imaging technology that can display the inner anatomic structure information of the object accurately and intuitively. In recent years, multi-detector CT develops rapidly, indicating CT entering a new period. With CT extensive clinical applications, we may encounter insufficient projection data reconstruction problem, e.g., in the cases of obstacle shading and shortening scanning time for the health consideration. In these cases, the acquired projection data will be in-complete. Studying how to accurately reconstruct CT images from insufficient projection data is of great significance for fundamental research and practical application.When the number of projections does not satisfy the data sufficiency condition, streak-like aliasing artifacts will be inevitable in CT images or the images are not complete reconstructed, using classical filtered back-projection (FBP) algorithms. So we need to seek new image reconstruction methods to address the insufficient projection data reconstruction problem. Compared with FBP methods, the advantage of CT iterative reconstruction methods are more obvious. The iterative reconstruction algorithm is able to generate tomographic images with higher quality when the projection data is noisy or insufficient. Hence, using iterative reconstruction algorithms to solve the problem will be a better technical route.At the same time, compressed sensing theory has proved that sparse signals can be accurately reconstructed from far less measurements than what is usually considered necessary according to the Shannon/Nyquist sampling theorem. It provides us potential direction for our problem. In this project, based on compressed sensing theory, we proposed and implemented a new category of CT iterative image reconstruction algorithm. We call them CSIR algorithms. The methods convert the problem to function minimization program under linear constraint, and consider how to construct different object function to make the result more accurate. Meanwhile, we used prior information to suppress artifacts and remedy lost data.Aiming at the reconstruction problem under insufficient projection data, we did lots of simulation research. Results showed that the proposed method could accurately reconstruct tomographic images from limited projections and the images are of high quality without streak-like aliasing artifacts. Traditional iterative algorithms or methods singly using compressed sensing may not achieve it. And the quantitive analyses for these results gave the same conclusions. So the proposed method is highly potential in clinical applications.