The accuracy of whole brain N-acetylaspartate quantification

Abstract

A non-localizing pulse sequence to quantify the total amount of N-acetylaspartate (NAA) in the whole brain (WBNAA) was introduced recently [Magn. Reson. Med. 40, 684–689 (1998)]. However, it is known that regional magnetic field inhomogeneities, ΔB₀s, arising from susceptibility differences at tissue interfaces, shift and broaden local resonances to outside the integration window, leading to an underestimation of the true amount of NAA in the entire brain. To quantify the upper limit of this loss, the whole-head proton MR spectrum (¹H-MRS) of the water was integrated over the same frequency width as the NAA. The ratio of this area/total-water-line was 75 ± 5% in 5 volunteers. The procedure was repeated with the brain-only water peak, obtained by summing signals only from voxels within that organ from a three-dimensional chemical-shift-imaging (3D CSI) set. It indicated that <10% of the water signal loss occurred in the brain. Therefore, by analogy, WBNAA accounts for >90% of that metabolite.

Introduction

The spatial extent of diffuse/multi-focal brain diseases is usually obtained from tissue contrast differences, elicited by enhancing agents and/or MRI pulse-sequences [1]. Unfortunately, conventional imaging is quite nonspecific with different pathologic processes, e.g., edema, demyelination, gliosis and necrosis appear similar [2], [3]. Furthermore, diffuse microscopic disease may be more pervasive than revealed by MRI [4]. To address these issues and provide additional information, MRI is sometimes augmented by localized ¹H-MRS, e.g., in studies of multiple sclerosis, traumatic brain injury, Alzheimer’s disease and AIDS [5], [6], [7], [8]. However, localized ¹H-MRS suffers from several limitations: (a) uncertainty in the reproducibility of the placement of a small volume-of-interest (VOI) in serial studies; (b) need for image-guidance of the VOI to visible pathologies; and (c) poor signal-to-noise-ratio (SNR) requiring prolonged acquisition.

To avoid these limitations, we recently proposed a ¹H-MRS method to evaluate changes in the brain as a whole, through quantification of the total amount of neuronal loss [9]. This is assessable from the level of NAA, an amino-acid derivative present exclusively in neuronal cells [10], [11] and which yields the most prominent peak in the in vivo brain’s ¹H spectrum [12], [13]. WBNAA acquisition does not suffer from the problem of serial VOI placement, image-guidance to visible pathologies is unnecessary and the SNR is >300:1, facilitating a quick, <3 min., acquisition [9].

However, it has long been known that magnetic field inhomogeneities, ΔB₀s, arising from susceptibility differences at tissue/air and tissue/bone interfaces broaden and shift resonance lines [14], [15], [16], [17]. These regional effects are especially pronounced in the frontal and temporal lobes above the sinuses, in the vicinity of the auditory canals, and near the skull [14], [16], [17]. They can be as large as 2.0 ppm over a 2 × 2×2 cm³ voxel [17] and cannot be removed with the 1st and 2nd order shims found in clinical MR imagers, as noted by Gruetter [18]. Since (a) quantification of NAA involves integrating its peak’s area and (b) it is desirable to keep the integration range, δ^NAA, narrow to avoid contamination from adjacent lines; it is likely that a fraction of the NAA resonances will be shifted outside δ^NAA, leading to area underestimation. This paper investigates and quantifies the extent of this error in WBNAA.