Sodium MR Imaging Detection of Mild Alzheimer Disease: Preliminary Study

Mellon EA, Pilkinton DT, Clark CM, Elliott MA, Witschey WR 2nd, Borthakur A, Reddy R.
AJNR Am J Neuroradiol 2009;30:978–84

Background: Essential for cellular homeostasis and electrochemical activity throughout the human body, sodium ions (²³Na) form a gradient due to differing intra- and extracellular concentration with [²³Na]_in being 10–15 mmol/l and [²³Na]_ex being 145 mmol/l, respectively. Cellular viability correlates to tissue [²³Na]. Brain pathologies such as tumors [1] or ischemia [2] alter the ²³Na gradient mostly due to increased [²³Na]_in. Using an MR coil resonant at 32.59 MHz (B₀ = 3 T), ²³Na magnetic resonance imaging (MRI) is able to assess total tissue sodium content.

Purpose: To test for differences in cerebral ²³Na tissue concentration between individuals suffering from Alzheimer’s disease (AD) and healthy controls.

Materials and Methods: Five patients with a clinical diagnosis of probable AD (Mini-Mental State Examination [MMSE] Score 18–28) and five age- and sex-matched healthy controls (MMSE Score 30) underwent ²³Na MRI on a 3-T Tim Trio Scanner (Siemens Medical Solutions, Erlangen, Germany). Using a custom-built head coil tuned to the ²³Na resonance of 32.59 MHz, a coronal sodium-spoiled fast gradient-echo (GRE) sequence was applied (TR [repetition time] 9.13 ms, TE [echo time] 2.96 ms, FA [flip angle] 60°, bandwidth 130 Hz/Px, FOV [field of view] 245 × 245 mm², voxel size 0.146 cm3). Image reconstruction was performed offline. To increase signal-to-noise ratio (SNR), data was K-space-filtered with three Gaussian functions. Subsequently, data was normalized to the top 10% of maximal signal intensities as measured in the lateral ventricles. Then, images were skull-stripped using the Brain Extraction Tool (BET, part of FMRIB’s software library FSL, http://www.fmrib.ox.ac.uk/fsl/) [3] and a total intracranial volume (TIV) was calculated. Furthermore, the hippocampus was manually segmented using ITK-SNAP (http://www.itksnap.org). ²³Na images were co-registered with corresponding T1-weighted images (SPM5, http://www.fil.ion.ucl.ac.uk/spm). Statistical computations, i.e., two-tailed t-tests and inference of Pearson’s correlation coefficients, were performed in Excel 2007 (Microsoft, Redmond, WA, USA) and SPSS (SPSS, Chicago, IL, USA).

Results: Mellon et al. have found (I) differences in overall normalized ²³Na signals between AD patients and healthy controls. Furthermore, (II) they describe a correlation between ²³Na signal and AD diagnosis (r² = 0.53) and (III) between ²³Na signal and TIV-normalized hippocampal volume (r²= 0.42). Moreover, (IV) Mellon et al. report an overall correlation between ²³Na signal of each hemisphere and corresponding unnormalized hippocampal volumes (r² = 0.50), a correlation between ²³Na signal of each hemisphere and unnormalized hippocampal volumes of (V) AD patients (r²= 0.40) and (VI) healthy controls (r² = 0.24).

Comment

For ²³Na MRI, Mellon et al. applied a GRE sequence at a TR of 9.13 ms and TE of 2.96 ms. There are two drawbacks arising from the choice of the applied sequence parameters. First, due to the TR of 9.13 ms images are strongly influenced by T1 effects. Second, corresponding to the biexponential T2 decay of tissue ²³Na signal characterized by a fast (T2_f 0.5–8 ms) and slow component (T2_s 15–30 ms), there is substantial loss of signal at a TE of 2.96 ms, for T2_f constitutes about 60% of the biexponential signal decay.

Per se, the sample size of the study by Mellon et al. (n = 5 AD patients and n = 5 healthy controls) is rather small for statistical analyses. Also, including all subjects (n = 10) into correlation analyses and even subdividing groups by introducing each hemisphere as factor of interest (n = 20; i.e., correlations between unnormalized hippocampal volume and each hemisphere of AD patients and healthy controls have been computed) surely is an invalid approach because of intrinsic correlations.

Mellon et al. render the applied ²³Na GRE sequence optimal for clinical routine because of a high SNR and image quality with minimal blurring. However, as stated above, the sequence used by Mellon et al. generates an intrinsic signal loss thus reducing the SNR. Moreover, as part of the postprocessing image data was filtered using three Gaussian functions. Filtering data also introduces blurring to the data. If filtering is necessary due to low SNR, a filter with a finite extent, e.g., a Hanning window [4], might have been a better choice. Other methods such as twisted projection imaging (TPI) [5], three-dimensional radial projection reconstruction (3D-RAD) [6], or a density-adapted three-dimensional radial projection technique (DA-3D-RAD) [7] using ultrashort TEs would be better options because higher SNR and less relaxation weighting can be achieved. Thus, the sequence used in the study of Mellon et al. is doubtlessly not optimal for clinical routine.

Due to intrinsic substance loss, AD patients exhibit sulcal widening. As a result, the ratio of brain tissue to cerebrospinal fluid (CSF) decreases. Thus, the signal CSF contributes to the overall ²³Na signal increases in patients suffering from cerebral atrophy. Therefore, the ²³Na GRE sequence as used by Mellon et al. yields rather unspecific information. Again, an inversion recovery ²³Na MR sequence [8] might provide more sensitive and specific results, for, in contrast to tissue-bound sodium, most of ²³Na in solution is suppressed.

In our opinion, results I–VI of the study by Mellon et al. are not definitely interpretable in the corresponding clinical context.

Providing information about biochemical tissue characteristics, ²³Na MRI is a very valuable technique especially for defining pathologies with altered ion gradients such as brain tumors or cerebral ischemia. However, the selection of sequence parameters is essential for highly sensitive and specific results.

References

1. Ouwerkerk R, Bleich KB, Gillen JS, Pomper MG, Bottomley PA. Tissue sodium concentration in human brain tumors as measured with ²³Na MR imaging. Radiology 2003;227:529–37.
2. Sandstede JJ, Hillenbrand H, Beer M, Pabst T, Butter F, Machann W, Bauer W, Hahn D, Neubauer S. Time course of ²³Na signal intensity after myocardial infarction in humans. Magn Reson Med 2004;52: 545–51.
3. Smith SM. Fast robust automated brain extraction. Hum Brain Mapp 2002;17:143–55.
4. Parker DL, Gullberg GT, Frederick PR. Gibbs artifact removal in magnetic resonance imaging. Med Phys 1987;14:640–5.
5. Boada FE, Christensen JD, Gillen JS, Thulborn KR. Three-dimensional projection imaging with half the number of projections. Magn Reson Med 1997;37:470–7.
6. Nielles-Vallespin S, Weber MA, Bock M, Bongers A, Speier P, Combs SE, Wohrle J, Lehmann-Horn F, Essig M, Schad LR. 3D radial projection technique with ultrashort echo times for sodium MRI: clinical applications in human brain and skeletal muscle. Magn Reson Med 2007;57:74–81.
7. Nagel A, Laun F, Weber MA, Matthies C, Semmler W, Schad L. Sodium-MRI using a density adapted 3D radial acquisition technique: in press.
8. Stobbe R, Beaulieu C. In vivo sodium magnetic resonance imaging of the human brain using soft inversion recovery fluid attenuation. Magn Reson Med 2005;54:1305–10.

(submitted July 16, 2009)

Armin Biller, Heidelberg, Germany