Introduction
Brain atrophy assessment is important in clinical and research neuroradiology settings. When evaluating brain atrophy, tissue loss is assumed when the peripheral and central cerebral spinal fluid (CSF) spaces are enlarged in relation to the intracranial volume [
1]. Brain atrophy is an important imaging sign in cerebral small vessel disease (SVD) [
2], which is the most common cause of vascular dementia and causes one-fifth of all strokes worldwide [
2,
3]. Also, brain atrophy predicts conversion to Alzheimer’s disease (AD) in patients with mild cognitive impairment (MCI) [
4].
Brain atrophy can be evaluated by means of quantitative brain volume measurements, or it can be visually scored by means of rating scales such as the global cortical atrophy (GCA) scale [
5] and the medial temporal lobe atrophy (MTA) scale [
6,
7]. Quantitative brain volume measurements are mainly used in research, because of their relative reliability and good sensitivity compared to visual scores. Typically, brain volume is expressed as a percentage of the intracranial volume. This requires both a T
1-weighted and a T
2-weighted sequence [
1] in which the T
1-weighted sequence is used to assess brain volume, while the T
2-weighted sequence is used for the measurement of intracranial volume [
1]. Currently, no image segmentation method is completely accurate, as brain tissue voxels containing CSF partial volume will bring in errors in volume measurements, even with high-resolution images [
8,
9]. Despite these inaccuracies, volumetric magnetic resonance imaging (MRI) can predict conversion to AD in accordance with the GCA and the MTA [
10,
11]. So far, no diagnostic threshold for computerized brain atrophy evaluation has been established. Moreover, the currently available segmentation techniques are not easily implemented in daily clinical care. Therefore, visual rating scales remain the gold standard in the clinical setting. Unfortunately, the GCA [
5] and the MTA scales [
6,
7] suffer from subjectivity. Inter-observer agreement for the GCA scale varies from poor to substantial (kappas of 0.34 and 0.7) [
5,
8], whereas the MTA scale seems more robust with inter-observer agreements between 0.82 and 0.9 [
12].
Recently, Qin [
9] suggested a simple approach to map and measure baseline CSF volume fraction [
9] as a surrogate for brain parenchymal volume, which could thus be used to evaluate brain atrophy. The approach is based on the long transverse relaxation rate of CSF (T
2,CSF) [
13] as compared to other tissue types [
14]. It allows for measuring the CSF volume and T
2,CSF simultaneously. Apart from its potential as a surrogate for brain atrophy measurements, the latter measurement (T
2,CSF) may reflect the CSF contents.
The main purpose of this study was to investigate the ability of this new, fast sequence (scan time of 1:11 min) to act as a surrogate for brain atrophy. A second purpose was to measure the T2 of the CSF and to investigate how it relates to the GCA and MTA scales.
Discussion
We demonstrated that a fast CSF MRI sequence with an imaging time of 1:11 minutes and fast post-processing without user interaction can be used to obtain the intracranial volume, the CSF volume and the T2 relaxation time constant of the CSF. The obtained volume measurements correlated well with the qualitative atrophy scales, which show that the CSF MRI sequence can be used to evaluate brain atrophy. We found a high correlation between the T2 of the CSF and the qualitative atrophy scales. The T2 of the CSF could thus be a marker for neurodegenerative diseases.
Over the last few years, an increasing number of research groups have started to use image segmentation tools, which are used to measure the brain parenchymal volume, instead of qualitative image atrophy scores to evaluate brain atrophy. In this paper, the percentage of CSF was used as a surrogate for brain parenchymal volume. The method that we used to obtain the CSF volume exploits the long T
2 of the CSF to isolate its signal. By fitting T
2 decay within each voxel, and comparing the fitted signal at TE = 0 [“S(0)”] to the signal of pure ventricular CSF voxels, the amount of CSF within each voxel is estimated. This yields accurate measurements of the CSF volume fraction within the voxel [
9], from which regional CSF volumes can be obtained. We limited ROI analyses to the peripheral subarachnoidal and the ventricular space. However, this could be extended to obtain measurements of, for instance, frontal, parietal or temporal CSF volume, and thus provide an assessment of regional atrophy. Regional atrophy assessment could potentially enable classification of dementia diseases based on their atrophy patterns [
20]. For instance, AD and frontotemporal lobe degeneration are both associated with medial temporal lobe atrophy [
21‐
24]. The combined presence of either temporal and posterior, or temporal and frontal atrophy may help to distinguish between both [
25‐
27]. Additionally, regional measurements could be helpful when differentiating ventriculomegaly or hydrocephalus from brain atrophy. Enlarged ventricular CSF volume without an increase in peripheral subarachnoidal CSF volume could guide the assessor towards ventriculomegaly or hydrocephalus instead of brain atrophy. It should be noted that the CSF MRI sequence cannot differentiate between gray or white matter loss, and as such, can only supplement conventional MRI sequences but not replace them.
In this paper, we compared the results of the brain parenchymal volume measurements obtained from our CSF MRI sequence to the results of a more standard computerized quantitative brain atrophy evaluation. We demonstrated a strong correlation between both brain parenchymal volume measurements. The linear analysis showed that our CSF MRI sequence estimates the ICV as systematically larger than the T1-based segmentation method by about 230 ml. This is probably due to a slight difference, in between both methods, in the exact (subvoxel) localization of the border between CSF and tissue. As well, small amounts of variation may be explained by partial volume effects in either the 3D-T1w sequence, caused by the inclusion of brain tissue in CSF voxels, or by partial volume effects in our CSF MRI sequence, which are discussed below. More recent segmentation software may perform even better, but the scope of our comparison was to compare to standard available segmentation software.
An earlier study demonstrated that over half of the variation in brain volume can be explained by ICV and a quarter of the variation in brain volume can be explained by gender [
19]. This was confirmed in our study, as we demonstrated a significantly different ICV in male compared to female subjects. The strength of the approach presented in this paper is that it accounts for this variation by expressing CSF volumes as a ratio to the ICV.
We found an increase in the T
2,CSF with increased visual ratings of global cortical and medial temporal lobe atrophy. Qin suggested that the regional variation in T
2,CSF is likely explained by variation in the composition of the CSF [
9]. The CSF composition in patients can be different due to both differences in O
2 partial pressure and differences in CSF protein content. Zaharchuk found a profound effect of oxygen partial pressure on the longitudinal relaxation rate (T
1) of the CSF [
28]. Even more so, they showed that inhalation of 100 % oxygen changed the T
1 of CSF, an effect which was most pronounced in the cortical sulci [
29]. This matches with our findings, which show that the T
2 of the peripheral subarachnoidal CSF has a stronger relationship with the GCA and MTA scale than the T
2 of the ventricular CSF. We hypothesize that subjects with more brain atrophy have a lower oxygen content of CSF. Small vessel disease could lead to decreased brain perfusion and thus decreased oxygen diffusion from the blood vessels towards the CSF. A decrease in oxygen content would lead to a longer T
2, which would correspond to our observations. Unfortunately, we were not able to measure the partial oxygen pressure of CSF in our subjects and thus to investigate this phenomenon. Further research with dedicated measurements of the partial oxygen pressure in the CSF are needed to investigate this. So far, the partial oxygen pressure of the CSF within this patient population has not been investigated, as the measurements are rather difficult to perform. Besides oxygen pressure, changes in the protein content of the CSF related to pathology should be investigated as a potential explanation for the regional variation in T
2 of the CSF. Moreover, though Qin [
9] suggested variation in composition of the CSF as the most likely source for variation in T
2 of the CSF, it may still be possible that the T
2 mapping sequence yields systematic errors in the estimated T
2. These might be related to partial volume effects and imperfections in the excitation and refocusing pulses (heterogeneity in the transmit RF field, B1). Thus, the reproducibility of the method and its robustness against B1 variation (due to the use of an MLEV T2 preparation scheme) should be established in future studies before firm conclusions on the origin of regional T
2 variation can be drawn.
In this paper, an MLEV scheme was selected based on the initial work of Qin [
9]. The advantage of the pre-pulse approach is that it can match resolution and coverage to other functional scans such ASL and functional MRI. This increases the abilities to perform partial volume corrections within these modalities, which is especially important in populations with varying degrees of atrophy.
Although the CSF MRI sequence seems promising, there are some limitations. One of them is partial volume effects, which could have potentially influenced our results and the found correlations. However, several observations demonstrate that the partial volume effects had no major influence on the results. First, the ventricles should be more or less free of partial volume effects, and we demonstrated that the correlations with the GCA and MTA scores hold true even when only the ventricular CSF volume or T2,CSF were evaluated. In addition, eroding the ventricular masks to smaller ROIs only had a minor influence on the T2,CSF values. Second, we investigated the influence of partial volume effects by comparing the results obtained with a low resolution (3 × 3 × 6 mm3) CSF MRI sequence to the results obtained with a high resolution CSF MRI sequence (1 × 1 × 4mm3) and found no significant difference between the peripheral subarachnoidal CSF volumes. However, the difference in between the T2,CSF, measured with both the low and high resolution CSF MRI sequence in the peripheral subarachnoidal space, was significant. As explained before, this may indicate that the T2 has a higher sensitivity to partial volume effects than the CSF volume measurements.
In conclusion, we demonstrated that a fast CSF MRI sequence has the potential to replace qualitative imaging atrophy scales and supplement conventional segmentation methods. The method is advantageous, as it requires little post-processing time and no user interaction. The T2,CSF should be investigated more thoroughly, as it could be a marker for neurodegenerative diseases.