CG-SENSE [1] is an iterative algorithm used in PMRI for MR image reconstruction from under-sampled data. One major limitation of CG-SENSE is the appropriate choice of the number of iterations required for good reconstruction results, fewer iterations may result in aliasing artifacts and too many iterations result in an increased noise level. This paper proposes the use of a correlation measure between the line profiles of the reconstructed images in the current and the previous iterations, as a stopping criterion for the CG-SENSE algorithm.Human cardiac data (in radial k-space) is acquired using 3 Tesla Skyra Siemens scanner at Case Western Reserve University, USA with 30 channel cardiac coils and 80 measurement frames (dimension 256×144×30×80) at TR=2.94ms with real-time, free-breathing and no ECG gating. The acquired data is under-sampled for different acceleration factors retrospectively. The under-sampled radial data is fed into CG-SENSE algorithm. A correlation [3] of the line-profile (the central line of the reconstructed image) between the current and the previous iteration is computed. An ideal value of this correlation (i.e 1) indicates that the iterations should stop because the most recent iteration has not improved the solution image. This paper uses a correlation value of 0.9950 as a stopping criterion. The reconstruction results of the proposed method are compared with the Bregman distance [2] stopping criterion, where Bregman distance between the current and the previous iterations was used as a stopping criterion.

Figure 1 shows the reconstructed cardiac images obtained using the line profile stopping criterion (proposed method). Figure 2 shows the reconstructed cardiac images at different acceleration factors (AF) using Bregman distance stopping criterion. Figure 3-a compares AP for both the stopping criteria and it shows that at AF= 4 the AP for Bregman distance method and line profile method is comparable whereas at higher acceleration factors (i.e. 8 and 12) line profile stopping criterion(proposed method) shows better results. Figure 3-b compares peak-signal-to- noise ratio for both the methods. The results show that the proposed method provides good PSNR at all acceleration factors.

The line profile correlation measure acts as an effective stopping criterion in CG-SENSE