Diffusion and Perfusion Magnetic Resonance Imaging: Application to Functional MRI
- R. P. Kennan, J. Zhong, J. C. Gore
Using Magnetic Resonance Imaging (MRI) the authors report the first observation of a transient decrease of the apparent diffusion coefficient (ADC) of water in the human brain visual cortex during activation by a black and white 8Hz-flickering checkerboard. The ADC drop was very small (below 1%), but significant and reproducible, and closely followed the time-course of the activation paradigm. Based on the known sensitivity of diffusion MRI to cell size in tissues and to optical imaging studies which have revealed the existence of shape changes in neurons and glial cells during activation, the observed ADC findings have been tentatively ascribed to a transient swelling of cortical cells. These preliminary results suggest a new approach to produce images of brain activation with MRI from signals directly associated to neuronal activation and not through changes of local blood flow. Introduction: Functional Magnetic Resonance Imaging (fMRI) has appeared as a very powerful tool for cognitive neurosciences, offering noninvasive access to the cortical and subcortical networks involved in sensorymotor and high order cognitive processes on an individual basis. FMRI relies on the common principle postulated at the end of the last century by Roy and Sherrington (1) that regional blood flow and metabolism are modulated by neuronal activity: FMRI provides quantitative or qualitative maps of changes in cerebral blood flow which are interpreted, on the basis of the above principle, in terms of regions being ‘activated’ (and sometimes ‘deactivated’) by sensorial, motor or cognitive tasks. The most common approach to fMRI has been the one based on ‘Blood Oxygen Level Dependent’ (BOLD) contrast (2-4). During brain activation, the large increase in blood flow which overcompensates a small increase in oxygen consumption results in a net decrease of the blood deoxyhemoglobin concentration which, in turns, as deoxyhemoglobin is paramagnetic, leads to a small, but measurable signal increase. Perfusion MRI techniques using contrast agents (5) or spin magnetic labelling methods (6) have also been used successfully. Still, the mechanisms and the origin of the neuronal activation/blood flow relationship are still unclear and the object of many fundamental research studies (see, for instance, review by Magistretti (7)). In any case, one should bear in mind that brain activation is seen with fMRI through a hemodynamic window and, thus, indirectly. Furthermore there is a delay of several seconds between the visible hemodynamic response and the onset of the neuronal response which intrinsically limits the temporal resolution of the method. Similarly, spatial resolution is limited, because the vessels at the origin of the current fMRI response feed or drain somewhat large territories compared to the activated neuronal assemblies. Ideally one should get direct access to small ensembles of neurons. So far, neuronal activity (action potentials) can only be recorded in animal models or preparations (beside recordings made intraoperatively in patients). Unfortunately, these studies are not easy to perform and analyze, even in animals. Furthermore, only a few ‘units’ can be recorded simultaneously in a given cortical region. One may dream of a non invasive imaging method allowing to bridge the gap between animal studies based on single unit recordings and macroscopic human studies. To do so, one must be able to directly visualize neuronal activity (and not remote vascular effects), to precisely identify the location of the active neuronal assemblies on the cortical ribbon. In this context several groups have explored new avenues to detect neuronal activation with MRI, such as the use of manganese tracers (8). On the other hand, diffusion MRI provides quantitative data on water molecular motion, a very sensitive marker of tissue structure at a microscopic scale, such as cell size or cell orientation in space (9). Preliminary data have suggested that water diffusion MRI could visualize changes in tissue microstructure which could arise during large, extraphysiological neuronal activation (10). Changes in water apparent diffusion coefficient (ADC) during neuronal activation would likely reflect transient microstructural changes of the neurons or the glial cells during activation. The possibility to observe such effects would have a tremendous impact, as they would be directly linked to neuronal events, to the opposite of the blood flow effects which are indirect and remote. However, this effect, if any, could be extremely small (a preliminary study performed in an isolated sciatic nerve did not show any water diffusion change (11)). The aim of this work was thus to investigate whether diffusion MRI would be sensitive enough to detect it. Methods: A visual stimulation study was conducted on 9 healthy volunteers (5 males and 4 females, ages 20 to 28 years). All participants gave their informed written consent. The study was approved by an Institutional Ethic Committee (#95-23). Diffusion and BOLD sensitized images performed on a whole body 1.5T MRI system (Signa, General Electric Medical Systems, Milwaukee, USA) equipped with an actively shielded whole-body magnetic field gradient set allowing up to 40 mT m. An echo planar imaging sequence, sensitized to diffusion by application of additional gradient pulses on either side of the refocusing RF pulse, was used to provide images with diffusion weighting (on all three gradient axes simultaneously, so as to maintain TE as short as possible, i.e. 72.3ms). This sequence was run by alternating two b values (b ~ 200 and 1443/1461 s/mm) to minimize effects of motion and eddy currents, at a rate of TR=1s for a total of 758 (or 418) repetitions. The b-values and the number of series and time points were set to optimize the signal to noise ratio in ADC images (σADC/ADC < 0.4 %). The performance of the sequence for diffusion measurement had been previously validated using a phantom containing isopropanol at 22 °C (12). An alternating black-and-white flickering checkerboard (8Hz) was used for blocks of 40 or 50s during 7 (or 4) ON/OFF cycles (Fig.1). Visual stimuli were delivered through a PC computer synchronized with fMRI acquisitions, using ‘EXPE5’ software (Pallier, Dupoux, & Jeannin, 1997). Three (or five) series of 8 axial slices across the calcarine fissure were acquired (64 matrix, 24cm FOV, 4mm/5mm thickness) resulting in more than 400 data points. BOLD fMRI data were also acquired with a T2*-weighted gradient EPI sequence (TR=1s, TE=60ms) using the same visual stimulation paradigm, but limited to 3 ON-OFF cycles. All experimental sessions occurred in complete darkness. Data preprocessing was performed with a home-written software that corrects for image distortion induced by eddy-currents in each series (13). Rigid motion effects were corrected across the three (or five) series. Statistical analysis was performed using SPM software (Friston, FIL, London (14)) for both BOLD and diffusion MRI time series. ADC maps were first calculated and smoothed (Kernel smoothing with FWHM = 8 mm × 8 mm × 8/10 mm). Activation-locked ADC changes were detected using a box-car model and t-test analysis. Clusters were meant significant when the associated p values where smaller than 5 10 (5 10 after volume correction for multiple comparisons). Statistical analysis was performed considering the whole brain volume and a subset corresponding to voxels found activated in the BOLD images (p <10). Results: In Figure 2 SPM-ADC maps of three adjacent slices in one subject are shown. A large cluster (84 voxels) of significantly decreased ADC is visible in the occipital cortex within an area also found activated with BOLD fMRI. The time course of the ADC in the most significant voxel of this cluster is shown as an example in Figure 3. Although the baseline is very noisy a drop in the ADC can be seen during each visual stimulation epoch. The average ADCs for each of 3 sequential experiments (series) for the baseline and activated conditions are shown in Fig.4. In average over the 3 runs the baseline ADC is 0.676 10mm2/s and significantly goes down to 0.664 10mm2/s during stimulation (1.79% drop). In order to better visualize the ADC change, the ADC time-course was averaged over the four activation blocks for each individual run and then smoothed out using a 5-points moving-average algorithm (Fig.5). The shape of the ADC timeresponse differs from the BOLD fMRI signal time-course, in particular by a slow return to baseline. However the moving average used for the ADC precludes any fine analysis of the ADC response time. Furthermore, the actual time resolution for the ADC images was 2s (instead of 1s for the BOLD images) due to the interleaved b1/b2 acquisition process. This pattern of ADC change was observed in the visual cortex of four participants, with cluster sizes varying between 12 and 84 voxels. The ADC drop (Table 1) was always very small (0.78% in average), but always significant except in one participant for which the ADC decrease remained under the statistical significance threshold after volume-correction (pcorrected ~ 0.1). Three subjects did not show any significant change in the ADC. In the remaining two subjects, strong paradigm correlated residual movement artefacts in diffusion-weighted images, prevented any analysis.