Prostate MR spectroscopic imaging (MRSI) with reduced acquisition time and without an endorectal coil could improve patient tolerance and throughput, and allow imaging of patients without rectum or with their prostate in normal position. The latter is important for radiation treatment planning but is difficult or impossible when an endorectal coil is used. In this work, we aim to demonstrate the following two important improvements in prostate MRSI: (1) a substantial reduction in the total scan time using a weighted acquisition scheme without SNR or protocol compromise, (2) a sorted singular value decomposition (SVD) method for combining spectra from a phased array surface coil. All experiments were performed on a 3T whole-body MRI scanner (GE Healthcare, Waukesha, WI) and using only one of the available phased array surface coils (an 8-channel or a 32-channel torso phased array by GE Healthcare, or a 16-channel “diaper” phased array coil by ScanMed, Omaha, NE). The scan-parameters for the prostate MRSI acquisition¹ were: TR/TE = 1300/85 ms, FOV = 110x55x55 mm, acquisition matrix = 12x8x8, voxel size = 6.9x6.9x6.9 mm, FID sampling points = 512, spectral bandwidth = 2 kHz. If a standard full k-space and symmetric sampling scheme were used, the total acquisition for the above protocol would be 16:44 min for a signal average of only 1. To improve the scan efficiency, we implemented an acquisition-weighted spectroscopic imaging technique with an odd symmetric sampling scheme, similar to that proposed by Pohmann et al, and by Scheenen et al^{2, 3}. Using the modified sequence for the same scan protocol, the total acquisition time was 9:42 min for a signal average of 4. For spectral postprocessing, we developed a sorted singular value decomposition (SVD) method for multi-channel multi-voxel spectral combination. Among the various techniques for combining multichannel MR spectra, SVD does not require a priori information on coil-sensitivity or estimating explicitly the complex coil weighting function4. In our sorted SVD method, the acquired signal from each receive channel is first Fourier-transformed along all three spatial dimensions. The resulting FID signal for each voxel is then sorted in a descending order according to an SNR index and assembled in an n x c matrix where n is the total number of the data points for each FID signal and c is the number of coils. The SNR used for sorting the signals of the different receive channels is obtained as the ratio of the water signal over the standard deviation of a pre-defined spectral region of noise for the spectra of each receive channel. Under the rank-one approximation, the n x c matrix for each voxel is decomposed using a standard SVD algorithm and the coil-combined spectrum is estimated as the Fourier transform of the left singular vector multiplied by the corresponding largest component of the diagonal matrix after the SVD decomposition⁴. To compare the improvement of our proposed modifications, we performed the SVD analysis of the same data without first sorting the FID signals.

Fig. 1 shows a comparison of the 32-channel coil-combined spectrum using the sorted SVD (red) and using SVD without sorting (green). The SNR of the water signal was 13.8 and 6.9, and the (choline+creatine)/citrate ratio was 0.46 and 0.37 for the two methods, respectively. The improvement shown was representative of the different spatial locations. The sorted SVD was found to always produce spectra with better quality and quantitation than SVD without sorting and therefore was used as our chosen method for processing.

Fig. 2 shows the spectrum of a patient from a region of biopsy proven cancer with an elevated (choline+creatine)/citrate ratio of 0.88. Fig. 3 shows the spectrum of another patient from a region of benign tissue with a (choline+creatine)/citrate ratio of 0.506. The spectrum in Fig. 2 was acquired using an 8-channel coil and the spectrum in Fig. 3 was acquired using the 16-channel “diaper” coil. Both were deemed to be of diagnostic quality according to a radiologist experienced in reading clinical endorectal coil prostate MRI and MRS.

Using an acquisition-weighted spectroscopic imaging sequence with an odd symmetric sampling scheme and a sorted SVD method for spectral processing of data from a multiple receive channels, we demonstrated that in vivo prostate MRSI can be performed at 3T in a substantially reduced scan time and without using an endorectal coil. Therefore, our proposed technique has the potential to help substantially expand the clinical use of prostate MRSI.