Case-PDM Optimized Random Acquisition in High Quality Compressed Sensing MR Image Reconstruction

INTRODUCTION Compresses Sensing (CS) has been recently introduced to speed MR imaging [1]. Among the parameters of CS, the undersampling scheme in k-space is vital to the success of the algorithm. To date, variable-density (VD) undersampling in phase-encodes is the most-used scheme in 2D Cartesian CS. However, VD schemes are based up experience rather than systematic optimization. In this paper, we use a perceptual difference model (Case-PDM) [2-4] to optimize VD schemes based on image quality. Experiments are conducted on a variety of brain data sets so as to ensure robustness across different MR systems, coils, pulse sequences, and subjects. METHODS To optimize VD, we parameterize the underlying probability density function (PDF) to consist of two equally weighted 1D Gaussians symmetric about the center of kspace, giving two parameters to optimize (σ and d). The PDF is used to control random acquisition for CS reconstruction, as described in Ref [1]. Case-PDM gives a scalar measure of image quality. On a training set of image data, we exhaustively evaluated a grid of PDF parameters, reconstructed images, and evaluated them using Case-PDM. To test the applicability of the optimized trajectory, we evaluated results on other test data sets. We used 5 brain data sets: D1 and D2 were transverse and saggital data acquired on a GE 3T using T1 FLAIR sequence; D3 and D4 were acquired on a Philips 3T system with an IR pulse sequence with different inversion times; D5 was acquired on a SIEMENS 1.5 T. (D1,D2,D3,D4) and (D5) were acquired with 8-channel and 4-channel head coils, respectively, from Invivo Corp, Gainesville, USA. 6 reduction factors were used. In total, 3600 images with a wide range of image quality were evaluated. RESULTS Figs. 1a and 2a show the comparison of the PDM scores of images reconstructed with different optimized trajectory and the experience-based trajectory [1] for different reduction factors. More low frequency signals and less high frequency signals are sampled by the optimized trajectory than those by the experience-based trajectory. When averaged across different reduction factors, the image quality improvement is 26% for the data set used for training, and 23% when the optimized trajectory was applied to other data sets. This demonstrates that acquisition trajectory parameters optimized from one training data set can be applied to other data sets of different slice/ subject/ pulse sequence/ scanner/ coil and maintain high image quality in CS reconstruction, as long as the two data sets share similar anatomical structures. The improvement of image quality and the application of the optimized trajectory can be further observed in Figs. 1 and 2. CONCLUSIONS The results demonstrate the applicability of Case-PDM on image quality evaluation of CS reconstruction. And we conclude that the PDM-optimized random acquisition trajectory can generate better CS reconstruction than the experience-based one. The optimized trajectory is robust across subjects and hardware configurations; that is, CS reconstructions with optimal acquisition trajectory maintain high image quality when the test dataset is different from the training dataset. Re-calibration is only necessary if the dataset has a different anatomical structure. ACKNOELEDGEMENT This work was supported under NIH grant R01 EB004070 and the Research Facilities Improvement Program Grant NIH C06RR12463-01. REFERENCES [1] M Lustig et al., MRM 2007 [2] KA Salem et al., Journal of Electronic Imaging, 2002 [3] D Huo et al., J Magn Reson 2008 [4] J Miao et al., Medical Physics 2008