Elektron spinn imaging! Accelerated 4D quantitative single point EPR imaging using model-based reconstruction Hyungseok Jang1, Shingo Matsumoto2, Nallathamby Devasahayam2, Sankaran Subramanian2, Jiachen Zhuo3, Murali C. Krishna2 andAlan B. McMillan1,* Article first published online: 6 MAY 2014 Author Information 1Department of Radiology, Wisconsin Institute for Medical Research, University of Wisconsin, Madison, Wisconsin, USA 2Radiation Biology Branch, Center for Cancer Research, National Cancer Institute, NIH, Bethesda, Maryland, USA 3Department of Diagnostic Radiology and Nuclear Medicine, University of Maryland, Baltimore, Maryland, USA DOI: 10.1002/mrm.25282 Purpose Electron paramagnetic resonance imaging has surfaced as a promising noninvasive imaging modality that is capable of imaging tissue oxygenation. Due to extremely short spin-spin relaxation times, electron paramagnetic resonance imaging benefits from single-point imaging and inherently suffers from limited spatial and temporal resolution, preventing localization of small hypoxic tissues and differentiation of hypoxia dynamics, making accelerated imaging a crucial issue. Methods In this study, methods for accelerated single-point imaging were developed by combining a bilateral k-space extrapolation technique with model-based reconstruction that benefits from dense sampling in the parameter domain (measurement of the T2* decay of a free induction delay). In bilateral kspace extrapolation, more k-space samples are obtained in a sparsely sampled region by bilaterally extrapolating data from temporally neighboring k-spaces. To improve the accuracy of T2* estimation, a principal component analysis-based method was implemented. Results In a computer simulation and a phantom experiment, the proposed methods showed its capability for reliable T2* estimation with high acceleration (8-fold, 15-fold, and 30-fold accelerations for 61×61×61, 95×95×95, and 127×127×127 matrix, respectively). Conclusion By applying bilateral k-space extrapolation and model-based reconstruction, improved scan times with higher spatial resolution can be achieved in the current single-point electron paramagnetic resonance imaging modality. Electron paramagnetic resonance imaging (EPRI) is a noninvasive imaging technique that measures the spatial distribution of unpaired electrons, akin to protons in magnetic resonance imaging (MRI). Owing to the recent development of biologically compatible spin probes [1-3], EPRI has emerged as a promising noninvasive imaging modality capable of dynamically and quantitatively imaging in vivo tissue oxygenation. However, due to extremely short spin–spin relaxation times, slice-selective imaging and conventional frequency encoding techniques are difficult to achieve and single-point imaging techniques are often utilized to improve image quality [4, 5]. In single-point EPRI (SP-EPRI), gradients remain constant during excitation, and data are acquired immediately after transmit dead time until no signal remains. Thus, SP-EPRI is rich in the spectral domain, but inherently suffers from reduced spatial and temporal resolution due to the time needed for a globally phase-encoded acquisition. Single-point imaging also exhibits a “zoom-in” effect due to the use of constant gradients, where k-space samples spread and objects enlarge (as field of view (FOV) decreases) at increasing phase encoding time delays. Recently, we proposed a method based on gridding termed k-space extrapolation (KSE) to maintain FOV across all phase encoding time delays and to improve the reliability of parameter estimation [6]. Although this method improves temporal resolution (by a factor of 3) by eliminating the need of multiple data acquisitions [7] required to secure multiple images with same FOV, further reduction in acquisition time is needed for single-point imaging. In MRI, a myriad of techniques have been proposed to accelerate imaging, such as parallel imaging [8-10], partial Fourier reconstruction [11] [also applied to SP-EPRI [12]], and compressed sensing reconstruction [13]. Among them, compressed sensing has recently surfaced as a promising method that can accelerate image acquisition by enabling high ratio of undersampling without a loss of image quality. Compressed sensing was first introduced in the area of signal processing and information theory [14, 15], which was based on the idea that signals can be reconstructed from highly reduced measurements if the signals show sparse representation. Recently, there have been many successful efforts using compressed sensing to medical imaging [16-20]. As medical images usually do not show sparse representation by themselves (except some special cases such as angiography), compressed sensing applications benefit from transform domain sparsity that is achieved by transformations such as the discrete wavelet transform. Unfortunately, the application of compressed sensing is difficult to use in SP-EPRI due to its small matrix size that inhibits transform domain sparsity. However, as SP-EPRI acquires abundant data in the parameter domain [measurement of the inline image decay of the free induction delay (FID)] and the inline image relaxation model is monoexponential and well known, SP-EPRI can benefit from model-based reconstruction techniques that simultaneously use k-p-space data in reconstructing images. In such model-based methods, an overcomplete dictionary or principal component analysis (PCA) can be used to sparsify the acquired data and improve inline image estimation [19, 21, 22]. In this study, we improved our previous KSE technique to add model-based reconstruction and further enhance spatial and temporal resolution in SP-EPRI. The improved, bilateral KSE (bi-KSE) allows more sample points to be secured in a target k-space by bilaterally extrapolating k-space samples from the neighboring k-spaces. In addition, a three zone sampling strategy was utilized for which a different sampling criterion was applied to each zone while taking advantage of the large degree of conjugate symmetry possible in EPRI. Model-based reconstruction using PCA [19] was implemented to further improve accuracy of inline image parameter estimation. During the reconstruction process, aliasing artifacts caused by undersampling are iteratively suppressed by promoting sparsity of principal coefficient (PC) maps in the discrete wavelet transform domain. The proposed reconstruction method allows significant improvement in the spatial resolution of single point EPRI. For example, if we acquire data with 61×61×61 gradient steps, the full sampling scheme will require a scanning time of approximately 75.7 min (226,981 points × 2000 averages × 10 μs TR), whereas 8-fold undersampling will enable imaging within approximately 9.5 min (28,373 points). Compared to methods that require multiple gradient acquisitions (typically 3), the proposed technique represents a 24-fold increase in temporal resolution. When using higher matrix size, we were able to achieve higher acceleration factors, for example R = 15 with 95×95×95 or R = 30 with 127×127×127 gradient steps. Nonetheless, these larger matrix sizes may not be realistic for SP-EPRI as a large number of sampled points are still required, for example 57,158 points and 68,279 points to achieve R = 15 with 95×95×95 and R = 30 with 127×127×127 matrix size, respectively, which are equivalent to 19.1 and 22.8 min of scanning time. Such scans might also operate at unrealistic signal to noise levels, despite our methods strong performance with low SNR data (SNRs of 5 for 127×127×127, 7 for 95×95×95, and 9 for 61×61×61 k-spaces, when measured at 1100 ns were utilized). Therefore, choosing a lower resolution (61×61×61) and moderate acceleration (R = 8) enables reasonable scanning time (
Posted on: Fri, 17 Oct 2014 16:40:19 +0000
Trending Topics
Recently Viewed Topics
© 2015