Patients were recruited from our hospital's Alzheimer's disease patient registry and research centre. At baseline, all patients met the criteria for a-MCI or AD. Patients were categorized in a special consultation with a neurologist and a neuropsychologist. Dementia and AD were diagnosed according to the DSM-IV and NINCDS-ADRDA criteria [²⁹], respectively. We selected patients with mild AD, corresponding to a Mini Mental State Examination (MMSE) [³⁰] score ≥ 20 out of 30. The diagnostic criteria for elderly patients with a-MCI were those published by Petersen et al. [²]: (i) a subjective memory complaint (i.e. reported by the patient or by his family), (ii) a documented, objective memory impairment adjusted for age and education, (iii) relatively normal general cognitive function, (iv) no dementia according to the DSM-IV criteria, and (v) generally preserved activities of daily living. Patients with a relevant neurological disease (stroke, tumour, etc.) or cerebrovascular risk factors (except for arterial hypertension controlled by medication) were not included in the study.

The third (NPH) group consisted of age-matched patients presenting with 2 or 3 of the Hakim triad symptoms (gait troubles were mandatory) with an insidious onset, no antecedent events such as head trauma, intracerebral haemorrhage or meningitis or other known cause of secondary hydrocephalus, ventricular enlargement on brain MRI (Evan's ratio > 0.3) and no macroscopic signs of obstruction to CSF flow. Based on these criteria and published guidelines [³¹], patients were classified as having "possible" NPH and the diagnosis was confirmed by specialist neurosurgical consultation and positive tap tests.

Hence, our patient populations consisted of 10 a-MCI patients (mean ± standard deviation (SD) age: 78 ± 7 years), 9 AD patients (mean ± SD age: 79 ± 5 years) and 13 NPH patients (mean ± SD age: 70 ± 6 years). The groups' demographic data and MMSE scores are summarized and compared in Table 1. All patients underwent the same PC-MRI imaging protocol. The patients were compared with a population (n = 12) of age-matched HEVs (mean ± SD age: 71 ± 9 years) with normal neurological and neuropsychological screening results (notably a normal MMSE score for educational level) and no active psychiatric, neurological or cognitive disorders.

The study protocol was approved by the local independent ethics committee. All participants had received an explanation of the study's objectives and procedures and provided their written informed consent to participation.

All MRI exams were performed with a standardized imaging protocol on a 1.5 or 3 Tesla (T) machine (Signa; General Electric Medical System, Milwaukee, WI, USA). Flow images were acquired with a 2 D fast cine PC-MRI pulse sequence with retrospective peripheral gating, so that the 32 analyzed frames covered the entire cardiac cycle (CC). The MRI parameters were as follows: 2 views per segment; flip angle: 25° for vascular flow and 20° for CSF flow; field-of-view (FOV): 14 × 14 mm²; matrix: 256 × 128; slice thickness: 5 mm; one excitation for the 3 T machine and two for the 1.5 T machine. Velocity (encoding) sensitization was set to 80 cm/s for the vessels, 10 or 20 cm/s for the aqueduct and 5 cm/s for the cervical subarachnoid space (SAS). A sagittal scout view was used as a localizer to select the anatomical levels for flow quantification. The selected acquisition planes were perpendicular to the presumed flow direction. Sections for each flow series are presented in Figure 1. The acquisition time for each flow series was about 90 s on the 3 T machine and 180 s on the 1.5 T machine with slight variations that depended on the participant's heart rate. The total, additional examination time for these CSF and blood flow investigations was around 10 min.

Data were analyzed using in-house image processing software http://www.tidam.fr. An optimized CSF and blood flow segmentation algorithm was used to automatically extract the region of interest (ROI) at each level. The ROIs (Figure 1) considered here were the right and left internal carotid arteries (ICAs), vertebral arteries (VAs), internal jugular veins (IJVs) and epidural veins (EVs) (for blood flows) and the aqueductal and cervical (C2-C3) areas (for CSF flows). In each ROI, flows were calculated for each of the 32 time frames in order to build a flow curve over the course of the CC. Flows in the ICA and VA on the left and right sides were summed to generate the total arterial cerebral blood flow (tCBF). We also calculated an arterial pulsatility index (API), which corresponded to the slope of the flow curve over time. We first measured the maximum and minimum flow amplitudes (F_max and F_min, respectively, expressed in ml/min) and their times of occurrence (T_max and T_min, respectively, expressed in milliseconds). The API index was then calculated as follows: API = (F_max - F_min)/(T_min - T_max)/2, and is represented in Figure 2.

We also used these parameters (F_max - F_min and T_min - T_max) to calculate the arterial volume pulsed at the beginning of systole, which is expressed as follows (in ml):

Arterial pulse volume = (F_max - F_min) × (T_min - T_max). Similarly, the cervical venous blood flow (VBF) was calculated by summing the venous flows in the left and right IJVs and EVs. A correction factor was applied to the amplitude of the internal jugular flow so that the mean total inflow arterial rate equalled the mean total outflow venous rate. The difference in total arterial and total venous flows over a CC corresponds to the arteriovenous (AV) curve, which is shown in Figure 2. After integration, the area under the curve represents the AV stroke volume (SV) and corresponds to the total blood volume mobilized in either caudal (output) or cranial (input) directions during the CC. In addition, we calculated the cervical arteriovenous delay (AVD), which corresponds to the time difference between the arterial systolic maximum flow peak (in the ICA) and the venous maximum peak (in the IJV). The AVD is expressed in milliseconds or as a percentage of the CC. Lastly, analysis of the CSF flow curves at aqueductal and cervical levels yielded the CSF SVs, which represent the oscillatory CSF volume displaced in both directions through the given ROI at each level [³²].

An additional file presents cerebral flows animation (Additional file 1. 3-D global animation of CSF and cerebral vascular flows). This video of an NPH patient was created by merging PC-MRI data sets with morphological MR images. It shows animated CSF/blood flows through the aqueduct of Sylvius, the C2-C3 subarachnoid spaces, the internal and vertebral carotid arteries and the jugular veins during the cardiac cycle.

Comparison of CSF and blood flow parameters in the four groups (AD, a-MCI and NPH patients and HEVs) was based on standard analysis of variance (ANOVA) for parametric data (tCBF, VBF, AVD, AV SV and cervical SV) and a Kruskal-Wallis one-way ANOVA on ranks for non-parametric data (API and aqueductal SV). When the ANOVA result achieved statistical significance, Bonferroni post-hoc analysis was used to identify the significantly different groups. The threshold for statistical significance was set to 0.05.

In patients with clinically defined a-MCI, arterial mean flows were significantly higher than in HEVs. The a-MCI patients also showed a much higher API (measuring the slope of the rise in arterial flow at the beginning of systole), which was consistent with the calculated arterial pulse volume (i.e. the arterial blood volume displaced in early systole). These changes in vascular dynamics may be correlated with the data on the prevalence of hypertension or other cardiovascular risk factors collected on inclusion. We observed mild diabetes in one a-MCI patient and similar proportions of arterial hypertension in the four study groups. Furthermore, all patients and healthy volunteers were only included in the study if their hypertension was medically controlled. Additionally, the a-MCI patients had a lower average heart rate than HEVs, although the difference was not statistically significant, contrasting with the higher arterial blood flow in a-MCI patients (Table 1). These results argue in favour of a change in the regulation of the arterial input (pumped by the heart) in a-MCI patients. An alternative explanation would relate to white matter vascular alterations. Although this parameter was not strictly quantified in the present study, no differences were readily apparent in a visual analysis.

In the absence of confounding parameters, our results suggest that a-MCI predemential states (even those without prominent leukoaraiosis) display a significantly elevated arterial cerebral inflow. These results are concordant with previous studies in which AD patients showed a high prevalence of risk factors for atherosclerosis, supporting the hypothesis that AD is "a vascular disease" [³⁸]. Anatomopathological studies have shown Aβ deposition in the leptomeningeal arteries in AD patients [³⁹] and capillary wall disruption is consistently found [⁴⁰]. Thus, it seems that increased artery pulsation results in regular dilation and then recoil, driving the interstitial fluid and proteins in the opposite direction to blood flow. The subsequent accumulation of Aβ in the brain's perivascular interstitial fluid drainage pathways could theoretically be responsible for the observed cortical deposits and clinical signs of AD.

The dynamic consequences of this theory have already been investigated in patients with AD. Two studies [⁴¹,⁴²] used transcranial Doppler ultrasound to evidence increased intracranial arterial pulsatility in AD patients (1.07 and 0.92), when compared with published reference data (a value of 0.83) [⁴¹]. Thus, it seems that the high extracranial arterial pulsatility demonstrated in our study in patients with neurodegenerative disease is transmitted (at least part) to the intracranial vascular compartment (where it can be detected in the basal arteries by using Doppler ultrasound). Elevated extracranial arterial pulsatility has also been suggested by MRI studies; Uftring [²¹] used mathematical transfer functions to compare the mechanical coupling of vascular and CSF flow oscillations in normal elderly subjects and AD patients. In both groups, there was a tendency for elevated vascular pulsations to cause elevated spinal cord oscillations and reduced subarachnoid CSF oscillations. The amplitude of these changes was greater in AD patients than in normal elderly subjects. Even though Barkhof [³³] did not find any changes in vascular or CSF flows in the elderly, we have previously shown [¹²] that normal ageing is associated with a significant fall in cerebral blood flow and in the pulsatility of the arterial input, as evidenced by a smoother systolic inflow peak than in young adults.

Therefore, our results suggest that a-MCI is associated with an abnormally high arterial pulsatility (independently of white matter abnormalities), which is probably due to degenerative changes in the arterial walls. These hyperdynamic arterial oscillations result in increased pulsatility at the aqueductal level. Indeed, the ventricular CSF SV was higher in a-MCI patients than in HEVs (although the difference did not achieve statistical significance); this is probably a reaction to arterial pulsatility and an attempt to dampen the increase in intracranial pressure.

Our results also revealed greater arterial pulsatility in AD patients, although the difference versus the other groups was not statistically significant. This change may be due to progression of arterial hyperdynamism during the transition from a-MCI to AD and a corresponding, steady reduction of cerebral compliance until a plateau is reached. The elevated arterial pulsatility is accompanied by relevant cerebral atrophy and a proportional decrease in energy needs and thus arterial perfusion. However, the establishment of this hypothesis is hampered by the small size of our groups and needs to be confirmed in larger populations.

The vascular results enabled us to discriminate between a-MCI and NPH patients, since the latter had significantly lower arterial flows and pulsatility indexes. In contrast, previously published data [⁴³] have not found arterial pulsation or the aqueductal SV to be of value in differentiating between AD and NPH patients. This discrepancy may be related to differences in selection criteria (i.e. pre-demential a-MCI in our study and patients with advanced AD in Bateman's study). Nevertheless, our results indicate that NPH and AD-like neurodegenerative diseases are probably caused by different etiological mechanisms. As mentioned above, we believe that arterial input may be the prime factor in the dynamic modifications seen in a-MCI; this would result in hyperdynamic functioning of the CSF and venous compartments and a moderate increase in ventricular CSF oscillation. In NPH patients, our results show low arterial pulsation and greatly elevated ventricular CSF oscillation.

In normal subjects, increased intracerebral volume usually caused by systolic arterial inflow, results in expansion of the arterial tree in two directions: (i) outward from the ventricles (with associated outward brain compression and CSF flush flow through the SASs) and (ii) inwards due to ventricle compression and aqueductal CSF flush flow. The elevated arterial inflow in a-MCI patients means that the arteries would essentially expand outward from the ventricles; this would result in greater expansion of the outer portion of the brain and subarachnoid CSF dampening of the pulse pressure. The inward compression of ventricles is moderate and so the aqueductal SV is not greatly elevated.

In NPH, the primary defect has been linked to a decrease in CSF oscillation [²⁷] (and thus compliance) in the cerebral SAS [²⁰] as a result of the compression of cortical venous drainage. Indeed, increased pressures in the cranial sinuses have been demonstrated in animal studies [⁴⁴]. Reduced CSF absorption in the venous dural sinuses then results in an inward compression of the ventricles and a greater aqueductal SV. The arterial flow reduction seems to be a secondary feature. If this disturbance were to persist, the ventricular system compliance would be overloaded and ventricle dilation would occur.

When comparing the four groups, NPH patients had a significant greater aqueductal SV than the HEVs and AD patients. Previous studies have suggested the usefulness of aqueductal oscillations as a helpful tool for diagnosis of NPH or for selection of patients who will benefit from shunting [²⁷,²⁸]. However, another study reported that aqueductal SV in AD falls between the values recorded in normal subjects and NPH patients and thus cannot be used to differentiate between the two groups of patients [⁴³]. These contradictory results may be due to differences in the respective study populations. Our AD patients were similar in terms of age and cognitive functions to those in the aforementioned study. Although cortical atrophy was not strictly calculated in our study, it appeared to be mild (and thus similar to the volume loss recorded in Bateman's study). The major difference concerns the aqueductal SV values in NPH patients, which differed significantly between the two studies (70 ± 50 in Bateman's study vs. 167 ± 89 in the present study). Our results are similar to those published by Luetmer et al. [³⁵] in what was probably a highly selected population. This may be due to different inclusion criteria; gait disorders were mandatory in NPH patients in our study, as this symptom is present in almost 80% of sufferers and is the first of the triad sign to appear. All patients were demented in Bateman's study and gait disorders were unnecessary if incontinence was present.

The difference in aqueductal SV between AD and NPH patients is probably due to different etiological factors. AD is associated with amyloid deposits, which can affect the small cerebral arteries and produce amyloid angiopathy [⁴⁰]. When combined with the greater prevalence of cardiovascular risk factors in AD patients [¹⁵,³⁸], amyloid angiopathy can increase the intrinsic stiffness of the cerebral artery walls and cause failure of the arterial pulse pressure dampening system. In AD, a primary arterial wall pathology may be responsible for greater outward brain expansion, compression of the subarachnoid spaces and thus an increase in spinal cord oscillations [²¹]. In contrast, it is thought (since publication of the "water hammer effect" theory by Greitz [³⁴]) that the primary etiology in NPH first affects the subarachnoid space and then the arteries, with a reduction in pulse pressure dampening.

Results are given for four groups: (i) healthy elderly volunteers (HEVs) and patients with (ii) amnesic mild cognitive impairment (a-MCI), (iii) Alzheimer's disease (AD) and (iv) normal pressure hydrocephalus (NPH). Values are expressed as the mean ± standard deviation. MMSE: Mini Mental State Examination; BPM: heart rate in beats per minute; tCBF: total cerebral arterial blood flow; API: arterial pulsatility index; VBF: venous blood flow; AV: arteriovenous; ml: millilitre; μl: microlitre; min: minute; CC: cardiac cycle.

The threshold for statistical significance was set to p < 0.05; NS: not significant.

HEVs: Healthy elderly volunteers; aMCI: patients with amnesic mild cognitive impairment; AD: patients with Alzheimer's disease; NPH: patients with normal pressure hydrocephalus; tCBF: total cerebral arterial blood flow; API: arterial pulsatility index; VBF: venous blood flow; AV: arteriovenous. P values in bold are statistically significant. P values on a grey background are close to statistical significance. NS: not significant.