An improved statistical iterative algorithm for sparse-view and limited-angle CT image reconstruction

Abstract

Because radiation is harmful to patients, it is important to reduce X-ray exposure in the clinic. For CT, reconstructions from sparse views or limited angle tomography are being used more frequently for low dose imaging. However, insufficient sampling data causes severe streak artifacts in images reconstructed using conventional methods. To solve this issue, various methods have recently been developed. In this paper, we improve a statistical iterative algorithm based on the minimization of the image total variation (TV) for sparse or limited projection views during CT image reconstruction. Considering the statistical nature of the projection data, the TV is performed under a penalized weighted least-squares (PWLS-TV) criterion. During implementation of the proposed method, the image reconstructed using the filtered back-projection (FBP) method is used as the initial value of the first iteration. Next, the feature refinement (FR) step is performed after each PWLS-TV iteration to extract the fine features lost in the TV minimization, which we refer to as 'PWLS-TV-FR'.

Figure 1

A flowchart of the proposed PWLS-TV-FR approach.

Figure 2

The head model for XCAT. Unless otherwise noted, all of the phantom images are shown in the same window [0.004 0.019].

Figure 3

Reconstructed image of the head model from 120 views using different methods. The images in the second row are partially enlarged views of the first row. (a) Reference (b) FBP (c) PWLS-TV (d) PWLS-TV-FR.

Figure 4

The residual image of the reconstructed results from Fig. 2. From left to right, the images are the residuals of results reconstructed by FBP, PWLS-TV and PWLS-TV-FR. The images are displayed in the window [−0.0015 0.0015].

Figure 5

RMSE and PSNR as a function of the total number of image updates. The RMSE function is shown on the left, and the PSNR function is shown on the right.

Figure 6

The left graph is profiles located at the pixel positions x from 200 to 300 and y = 260 of different results in Fig. 3. The graph on the right is the residual value corresponding to the left graph.

Figure 7

Image quality metrics.

Figure 8

The anesthetized mouse data reconstructions from 225 views by different methods. From left to right, the images of the first row are the results obtained using FBP, PWLS-TV and PWLS-TV-FR method. The images of the second row are the partial enlarged view of the first row.

Figure 9

The mouse bone sample data collected over a 197.88° range reconstructions by different methods. From left to right, the images of the first row are the results obtained using FBP, PWLS-TV and PWLS-TV-FR method. The images of the second row are the partial enlarged view of the first row.