No abstract

Key words: spectroscopic imaging; steady state free precession; rat brain; 1-H Simultaneous measurements of spectra in a large array of voxels can be performed by classical spectroscopic imaging (SI) which uses different phase encodings between RF excitation and signal detection (1,2). However, while the signal-to-noise ratio (SNR) per unit measurement time (SNR t ) is high, the minimum total measurement time (T min ) may be too long for certain in vivo applications, e.g., measurements with 3D spatial resolution or time-resolved measurements (3). Therefore, several modifications of classical SI (4 -6) as well as a large number of fast SI methods have been proposed to decrease T min , allowing spectroscopic measurements with 2D or 3D spatial resolution and large matrix size within minutes or even seconds. However, since the SNR t is crucial for in vivo studies, the higher speed in k-space sampling must not be achieved by a considerable decrease in the SNR t .Many fast SI techniques (7-16) are spectroscopic variants of fast MRI methods. Thus, fast SI methods based on echo planar imaging (EPI) (17), fast low-angle shot imaging (FLASH) (18), rapid acquisition by relaxation enhancement (RARE) (19), ultrafast low-angle RARE (U-FLARE) (20), BURST imaging (21), spiral imaging (22), gradient and spin echo imaging (GRASE) (23), or other fast MRI techniques have been proposed and applied in SI studies on humans or animals.Recently, fast imaging sequences using steady-state free precession (SSFP) methods have attracted increasing interest. While the basic ideas of fast SSFP imaging sequences were published more than 15 years ago (24 -28), only the improved hardware available on modern imaging systems allowed use of the specific image contrast and the high SNR t of SSFP sequences (29). A large number of SSFP pulse sequences with even more acronyms have been proposed which use either the FID-like signal S 1 , the echolike signal S 2 , or both signals in different ways. In particular, the TRUE-FISP sequence (24) in which S 1 and S 2 are coherently superimposed has been successfully used.Considering the high SNR t reported for SSFP imaging sequences, the potential of spectroscopic variants for fast SI should be evaluated. Recently, Speck et al. (30) used the TRUE-FISP sequence for fast 31 P SI. It was shown that the phosphocreatine signal can be imaged with high SNR in vivo. However, other signals were considerably reduced because of the strong signal dependence on the offset frequency.SSFP sequences may not have yet been considered for fast proton SI because of several reasons. First, the strong dependence of the signal intensity on the relaxation times T 1 and T 2 will make a quantification of metabolites difficult. Second, because the predominantly used TRUE-FISP sequence shows strong off-resonance effects, one might expect problems for the simultaneous detection of several metabolite signals. Third, the delay between subsequent RF pulses must not be too short to achieve a sufficient phase dispersion for the different chemi...

No abstract

Recently, new methods for fast (1)H spectroscopic imaging based on the condition of steady state free precession (SSFP) were introduced to achieve a high signal-to-noise ratio at short minimum measurement times. In this work, a major improvement is presented to overcome a crucial drawback in some of the former sequences: the lack of spatial selectivity. Good spectral selectivity at very high sampling efficiency can be achieved by using spectral-spatial RF pulses, and combined with localised shimming. Results are shown from both phantom experiments and in vivo studies on the rat brain acquired at 4.7 T.

Abstract. We show the impacts of various signal preprocessing techniques -dimensionality reduction and transformations -for high-resolution NMR spectra on the classification accuracy of different breast cancer tissue. Our results show that some preprocessing algorithms that are widely used nowadays will not reduce the data dimensionality in an information-preserving way: the classification accuracy drops. Besides showing the most successful preprocessing steps, we can report excellent results on a challenging classification problem.