CSF fraction calculation for single voxel spectroscopy: comparison of water signal T2 biexponential fitting and image segmentation in a pediatric population
Frances C Robertson1, Martha J Holmes1, Francesca Little2, Mark F Cotton3, Els Dobbels3, Andre JW van der Kouwe4,5, Barbara Laughton3, and Ernesta M Meintjes1

1Department of Human Biology, University of Cape Town, Cape Town, South Africa, 2Department of Statistical Sciences, University of Cape Town, Cape Town, South Africa, 3Department of Paediatrics & Child Health, Stellenbosch University, Stellenbosch, South Africa, 4A.A. Martinos Centre for Biomedical Imaging, Massachusetts General Hospital, Charlestown, MA, United States, 5Department of Radiology, Harvard Medical School, Boston, MA, United States

Synopsis

For partial volume correction in 1H-MRS the voxel fraction of brain matter (BM) and cerebral spinal fluid (CSF) can be calculated via biexponential fitting of T2 relaxation of the unsuppressed water signal or via segmentation of a high-resolution structural image. We compared voxel CSF percentages obtained using these two methods and investigated whether discrepancies could be explained by head movement between voxel positioning and MRS acquisition. Subjects with large differences in CSF% between methods tended to show greater displacement than those with no difference between methods. Inconsistencies may be due to segmentation inaccuracy in particular regions or subject motion.

Purpose

Because 1H-MRS requires large voxel volumes, a single voxel may contain a mixture of tissue types making partial volume correction necessary. Partial volumes of brain matter (BM) and cerebral spinal fluid (CSF) can be calculated via biexponential fitting of T2 relaxation of the unsuppressed water signal to obtain amplitudes of BM and CSF components at TE=0ms(1). Alternatively, segmentation of a high-resolution structural image can be used to determine gray matter (GM), white matter (WM) and CSF fractions within the voxel, assuming no head movement between voxel positioning and single voxel spectroscopy (SVS) acquisition. We aimed to compare these two methods and to determine whether large differences in CSF fraction between the two methods in 7-year-old children corresponded to substantial subject movement between acquisitions.

Methods

Participants were 113 7-year-old children from a neuroimaging follow-on study of the Children with HIV Early Antiretroviral Therapy (CHER) trial.

Scanning was performed on a 3T Allegra (Siemens, Erlangen, Germany). The protocol included a high-resolution T1-weighted 3D EPI-navigated(2) multiecho magnetisation prepared rapid gradient echo (MEMPRAGE)(3) (FOV 224x224 mm2, TR 2530 ms, TI 1160 ms, TE’s 1.53/3.19/4.86/6.53 ms, bandwidth 650Hz/px, 144 sagittal slices, 1.3x1.0x1.0 mm3), followed by two 3D EPI-navigated DTI acquisitions(4). SVS was obtained in the basal ganglia (BG), midfrontal gray matter (MFGM) and peritrigonal white matter (PWM), using an EPI volumetric navigated point-resolved spectroscopy (PRESS) sequence with real-time first-order shim and motion correction(5) (TR 2000 ms, TE 30 ms, 64 measurements, vector size 512, spectral bandwidth 1000 Hz). Water unsuppressed ¹H MRS measurements were acquired at TE’s of 30ms, 50ms, 75ms, 100ms, 144ms, 500ms and 1000ms.

The voxel water signal S(TE) was quantified using LCModel and modeled in Matlab as a biexponential function of TE as follows:

$$S(TE)=S_{CSF}.exp\left(\frac{-TE}{T2_{CSF}}\right)\left[1-exp\left(-TR\frac{-TR}{T1_{CSF}}\right)\right]+S_{BM}.exp\left(\frac{-TE}{T2_{BM}}\right)\left[1-exp\left(-TR\frac{-TR}{T1_{BM}}\right)\right]$$

where SCSF and SBM are the fitted signals from CSF and BM, T2CSF and T2BM are the fitted T2 relaxation times, and T1CSF and T1BM are T1 times(1). The component with the longer T2 was assigned to CSF.

Percentage CSF (PVCSF) in the voxel was estimated as:

$$PV_{CSF}=\frac{S_{CSF}}{S_{CSF}+S_{BM}\times0.75}\times 100$$

where SCSF and SBM are the signals from CSF and BM at TE=0, assuming that the signal from BM accounts for 75% of tissue volume, since BM is 75% water(1).

For each subject and region, the T1-weighted image was used to segment the SVS voxel into GM, WM and CSF in SPM12.

For cases where the difference between segmentation and biexponential methods was greater than 10% CSF, the displacement between the initial T1 image and the start of each SVS acquisition was calculated from motion parameters obtained by aligning the EPI navigator images from the preceding DTI and PRESS acquisitions in SPM12. For comparison, displacements were also calculated for 7 subjects where there was no difference in between the two methods.

Results

BG: PVCSF determined via segmentation was almost zero (<1%) for all but 2 subjects. Similarly, using the water signal T2 relaxation method a monoexponential fit (PVCSF= 0), was appropriate in more than half of the subjects, although the upper range of PVCSF calculated by this method was 95% (table 1).

PWM: PVCSF from the segmentation method was <1% in all but 2 cases. Again the range was larger using the T2 relaxation method.

MFGM: There was a range of PVCSF for both methods (figure 1), with a larger range for the T2 relaxation method.

The cumulative histogram of differences in PVCSF between the two methods (figure 2), shows differences of less than 1% CSF for 3/4 of the subjects in the BG but for only 1/3 of subjects in the PWM and 1/4 in the MFGM.

Subjects with PVCSF differences >10% CSF between methods showed a tendency for greater displacement between the T1 and SVS scans than those with no difference in PVCSF between methods (mean displacement difference: BG 2.8mm, p=0.04; PWM 2.5mm, p=0.07; MFGM 1.43mm, p=0.2).

Discussion

Agreement between the methods was best in the BG, since most PVCSF were 0 for both methods. In the MFGM a number of points fell around the y=x line, indicating fair agreement for certain subjects. In the PWM, the biexponential method estimated consistently larger PVCSF, since most of the CSF fractions calculated by the segmentation method were close to zero. In each region, there were some large discrepancies between methods that could not be fully explained by motion.

Conclusion

There is most consistency between methods in regions where the PVCSF is small, and less consistency where there are larger amounts of CSF. Inconsistencies may be due to segmentation inaccuracy in particular regions or subject motion.

Acknowledgements

Support for this study was provided by NRF/DST South African Research Chairs Initiative; US National Institute of Allergy and Infectious Diseases (NIAID) through the CIPRA network, Grant U19 AI53217; NIH grants R01HD071664 and R21MH096559; NRF grant CPR20110614000019421, and the Medical Research Council (MRC). We thank the CUBIC radiographers (Marie-Louise de Villiers and Nailah Maroof), our research staff (Thandiwe Hamana and Rosy Khethelo), and Shabir A. Madhi for access to control participants on the CIPRA-SA04 trial.

References

(1) Ernst T, Kreis R, Ross BD. Absolute Quantitation of Water and Metabolites in the Human Brain. I. Compartments and Water. Journal of Magnetic Resonance, Series B. 1993;102(1):1–8

(2) Tisdall MD, Hess AT, et al. Volumetric navigators for prospective motion correction and selective reacquisition in neuroanatomical MRI. Magn Reson Med. 2012;68(2):389-99.

(3) van der Kouwe AJ, Benner T, Salat DH, Fischl B. Brain morphometry with multiecho MPRAGE. Neuroimage. 2008;40(2):559-69.

(4) Alhamud A, Tisdall MD, Hess AT, Hasan KM, Meintjes EM, van der Kouwe AJ. Volumetric navigators for real-time motion correction in diffusion tensor imaging. Magn Reson Med. 2012;68(4):1097-108.

(5) Hess AT, Tisdall MD, Andronesi OC, Meintjes EM, van der Kouwe AJ. Real-time motion and B0 corrected single voxel spectroscopy using volumetric navigators. Magn Reson Med. 2011;66(2):314-23

Figures

Table 1: Descriptive statistics of CSF fractions for 3 different regions, obtained via image segmentation and biexponential fit to T2 decay of water signal.

Figure 1: CSF % in midfrontal gray matter determined using biexponential fitting (x axis) and segmentation (y axis) methods. Blue line is regression line with gray showing confidence interval, black line shows slope = 1, intercept = 0 illustrating line of no difference between methods. In this region, all CSF % measures of 0 from the biexponential method are non-zero in the segmentation method, which influences the apparent lack of agreement between methods.

Figure 2: Cumulative frequency of absolute difference in voxel CSF % between T2 biexponential and segmentation methods in 3 brain regions.



Proc. Intl. Soc. Mag. Reson. Med. 24 (2016)
4001