brain

 

 

 

 

 

 

 

 

 

 

Our research interest is the study of brain using magnetic resonance imaging, in particular its applications to functional neuroimaging and to the study of pathophysiology and early diagnosis for brain diseases.

Quantitative estimation of blood oxygenation and hematocrit in brain

The venous oxygenation (Yv) in the brain, has already been the interests of many MRI studies, because non-invasive and accurate measurement of Yv is not only important in terms of the fMRI signal, but also has directly clinical implications regarding the metabolic state of the brain. However, a robust method to measure this parameter has not been established. One of the main hurdle is that, in order to quantify the absolute value of blood oxygenation, one needs to isolate the blood signal (i.e. not having any partial volume from tissue or CSF). To solve this problem, we have recently applied spin labeling technique on the venous side (Fig. 1) (instead of the arterial side as the conventional ASL does) and performed paired-subtractions just like the ASL data processing (Fig. 2). Then the subtracted image only contains the venous blood signal. This signal by itself is not particularly interesting per se because it reflects the venous outflow but not perfusion. However, the T2 signal decay of this signal is interesting because the T2 relaxation time can be converted to Yv with a calibration plot. Another technical issue, the outflow effect of blood on the T2 measurement, is accounted for by using a non-selective CPMG-T2 preparation rather than conventional T2 weighting and possible imperfection in the 180° pulses is minimized by using a composite refocusing pulse as well as a MLEV phase cycling strategy (Brittain et al., 1995).

fig1

Fig. 1: Geometric locations of the labeling slab and the image slice in TRUST MRI for venous oxygen fraction measurement. The spins above the imaging slice are labeled, and, for the case of sagittal sinus, the venous blood inside will flow downward and enter the imaging slice. In the “control” scan, the spins above the imaging slice will not be labeled. Therefore, by performing a subtraction of the two images, only flowing signal remains.

Fig. 2a and b shows the control and label images at different effective TEs. Fig. 2c shows the corresponding subtracted images. Note that only large venous vessels have discernable signal intensities due to relatively large flow velocities. An ROI was drawn in the sagittal sinus area (Fig. 2d) and the signals from 4 voxels with highest amplitudes were averaged. It is important to point out that the ROI selection does not significantly affect the fitted T2 values, as does the choice for number of voxels. We have tested using 3-8 voxels and the results were virtually the same. The reason is that, even though the S0 (in Fig. 2e) is different in different voxels, the decay constant is roughly the same. Fig. 2e shows the experimental data and fitting results of the subtracted signal as a function of effective TE. Excellent fitting is achieved without contamination of other tissue signals and outflow effect. The resulting CPMG-T2 values are compared to a calibration curve (Fig. 2f) obtained from in vitro blood measurements at similar conditions (3T, =10ms, Hct=0.44) to estimate Yv in the human subjects. Table 4 summarizes the CPMG-T2 and Yv values for all subjects.

1a 3b4c

5d 6e 2f

Fig. 2: TRUST MRI data. (a) Control images at different effective TE (the effective TE is controlled by the number of refocusing pulses). (b) Label images. (c) Subtracted images (control-label). (d) ROI selection in sagittal. (e) Signal decay as a function of effective TE, from which the T2 value can be fitted. (f) Calibration curve to convert blood T2 to blood oxygen saturation fraction.

Functional brain mapping using cerebral-blood-volume based MRI contrast

fig5

Fig. 5: Visual activation maps (p<0.005) for BOLD-fMRI (a) and VASO-fMRI (b). Color bar indicates amplitude of fractional signal changes (positive for BOLD and negative for VASO). Activation data are overlaid on a high-resolution anatomical image. Spatially similar activation patterns are detected by both fMRI methods, although BOLD has more activated voxels than VASO. (c) and (d): average time-courses (n = 5) of the activated voxels for BOLD- and VASO-fMRI, respectively. Solid bars below the plots denote the timing of the stimulus paradigm. At this low spatial resolution (4x4x10 mm3), a 1.5% signal increase and a 0.7% signal decrease are seen for BOLD and VASO, respectively.

While blood oxygenation based and perfusion based fMRI methods have been widely studied by many laboratories, CBV-based fMRI method is not well established. Recently, we introduced a new fMRI technique that is sensitive to blood volume changes during neuronal activity (Lu et al., 2003). Such vascular-space-occupancy (VASO)-dependent fMRI uses a nonselective inversion pulse in combination with an optimal TI to eliminate blood signals. The remaining tissue signal is determined by the amount of extravascular water protons in a voxel, and is therefore directly related to CBV. Figure 5 shows BOLD-fMRI and VASO-fMRI activation maps obtained during visual stimulation. For both methods, clear activation can be seen in the part of the primary visual cortex corresponding to the central visual field. BOLD data show a larger number of activated voxels than VASO fMRI (p<0.005), namely an average (n=5) of 118 ± 16 versus 57 ± 22, respectively. This difference is partly attributed to the lower signal-to-noise ratio in the VASO case. Also, large venous vessels are known to cause extra activation areas in BOLD fMRI, which should not occur for the microvascular-based VASO-fMRI. The time-courses of the activated voxels are shown in Figs. 5c (BOLD) and 5d (VASO). The BOLD signal increases by 1.5% during stimulation, while the VASO fMRI signal decreases by 0.7%.

The VASO fMRI signal is related to CBV by:

3                                                       (7)

in which  is the water density of microvascular blood (or tissue) in ml water/ml blood (or tissue) and is well established in literature ml water/ml parenchyma, ml water/ml blood (Herscovitch and Raichle, 1985). As can be seen in the equation, the resting state CBV will need to be assumed or measured in order to calculate the CBV from the VASO signal.

Fig. 6: (a) Representative color map coded by the shift value at which maximum cc with the stimulation paradigm is achieved. Red-yellow color corresponds to the stimulation to the horizontal meridian in the visual field and blue-green color corresponds to vertical meridian. (b) VASO (left) and BOLD (right) retinotopic maps overlaid on an inflated cortical surface of a left hemisphere. Top row: medial view. Bottom row: posterior view.

We have further developed a sequence for multi-slice acquisition of the VASO images, in which the blood signal is continuously nulled by using a series of global inversion pulses (Lu et al., 2004b). We have demonstrated that VASO fMRI can be used to study the retinotopic mapping of human visual cortex (Lu et al., 2005b).

Determination of absolute cerebral-blood-volume in humans

Cerebral blood volume (CBV), defined as ml of blood per 100 ml of brain tissue, is an important measure in understanding brain physiology and pathophysiology. Many pathological conditions are associated with abnormal CBV values, including acute stroke and Alzheimer’s Disease. In this project, a novel approach will be developed for the accurate measurement of absolute CBV using Vascular-Space-Occupancy (VASO) MRI, a blood-nulling pulse sequence, in combination with the T1 shortening effect of the contrast agent Gd-DTPA. Two VASO images are acquired before and after contrast agent injection, resulting in a difference image that can be used to determine CBV. The key novelty of the proposed approach is that the estimation of CBV does not require knowledge or assumptions about vascular morphology. This is expected to provide a more accurate estimation of CBV compared with existing methods. The imaging protocol will be optimized by investigating various confounding factors that may cause error in the CBV estimation, including contrast agent concentration, MR receiver coil sensitivity profile, transverse relaxation effect of the contrast agent, effect of water exchange in the capillary bed, and effect of leakage in the blood-brain-barrier. The MRI CBV results will be validated using PET CBV measurements.

Fig. 3: (a) Timing diagram for the different experiments performed in this study. The scan durations for the experiments are also listed. (b) Schematic diagram of the multi-slice VASO pulse sequence. A global inversion is followed by an optimal TI to null the pre-contrast blood signal. Each slice acquisition takes 30 ms.

Fig. 4: (a) pre-contrast VASO image. Arrows indicate the sagittal sinus. (b) post-contrast VASO image. The image intensities are MR signals in arbitrary unit. (c) Reference image from a slice containing large CSF regions. The black box indicates the ROI used to calculate the normalizing factor. (d) Multi-slice aCBV map in ml of blood/100ml of brain. Display window is truncated at 20 to show contrast between gray and white matter.

Temporal dynamics of vascular and metabolic responses during physiological challenges

Figure 13 shows the activation maps using VASO, BOLD and ASL fMRI, respectively. We then measured the temporal characteristics of cerebral blood flow, volume, and oxygenation in human subjects (n = 8) before, during, and after visual stimulation. Fig. 14 shows the

fig13

Fig. 13: Activation maps (p<0.005) using VASO (a), BOLD (b), and ASL (c) fMRI during visual stimulation. Color bar indicates the magnitude of the cross-correlation coefficient. Maps are overlaid on VASO-EPI images (64x64).

time-courses of the hemodynamic responses for these physiological quantities using a measurement resolution of 2 seconds. When selecting all activated voxels for

fig

Fig. 14: Temporal dependence of physiological responses due to visual stimulation (n = 8, error bar = SEM). (A) Cerebral blood oxygenation (BOLD), flow and volume responses when selecting all activated voxels. A post-stimulus BOLD undershoot (at time > 38s) as well as a post-stimulus CBV delay are apparent. (B) When selecting only the subset of voxels activated in all three methodologies (overlapped voxels), negligible CBV and CBF changes are found after BOLD signal crosses the baseline (time ~38s), while a very large and prolonged BOLD undershoot remains. (C) Stimulation response curves for OEF and ∆CMRO2/CMRO2. The post-stimulus increase in OEF reflects dissociation between tissue oxygen use and vascular response, which is incompatible with a consequential coupling between flow and oxygen metabolism during synaptic activity. The brown and red curves were obtained using only the BOLD response, at the time when no CBV and CBF changes were detectable. This removes the effect of noise from the CBV and CBF curves on these calculated parameters. (D) Normalized initial changes (first 10 s) show that CMRO2 changes precede simultaneous CBV and CBF changes, which again precede the BOLD effect.

each of the different fMRI approaches (Fig. 14a), it is found that the responses for oxygenation and flow return to baseline within approximately 8-10 s after stimulus cessation, while the return of the blood volume response takes 5-10 seconds longer. The positive BOLD response is followed by a post-stimulus undershoot that has been the topic of much debate (Buxton et al., 1998; Mandeville et al., 1999). This oxygenation undershoot and the delayed CBV return have also been observed in rat studies, leading to the proposal that delayed venous/venular compliance may explain this phenomenon (Buxton et al., 1998; Mandeville et al., 1999). However, it can be seen that the BOLD undershoot lasts much longer than the prolonged blood volume increase, as long as 30s after the stimulation is stopped, and thus cannot be explained solely by such a delayed CBV return. This discrepancy becomes even more pronounced when focusing attention on the subset of voxels that show activation in all of the three methodologies (Fig. 14b). The time-courses show a very clear temporal mismatch between the post-stimulus BOLD undershoot and the CBF and CBV responses, indicating that this undershoot cannot be due to vascular dilatation and/or flow changes. Because no blood volume changes are evident during the post-stimulus BOLD undershoot in voxels activated in all three methodologies (Fig. 14b), this negative BOLD effect must be due to an increase in OEF. In the absence of flow changes, this undershoot effect can only be interpreted as a continued post-stimulus elevation in  (Eq. 6).

When calculating OEF and the relative changes in  with respect to baseline, the results (Fig. 14c) show that  is still elevated by more than 10% at a time when CBF and CBV have returned to baseline, and returns to normal levels over a period of about 30s. Such a post-stimulus elevation in oxygen metabolism is similar to the so-called “initial dip” (Ernst and Hennig, 1994; Menon et al., 1995) effect at the early onset of activation in that it reflects a  increase at a time that CBF is not increased. However, the effect of the initial dip is quickly counteracted by the effect of increasing CBF (within 1-2 s), and it is unclear whether this hyperemia is caused by the increase in energy use or by other causes. On the other hand, the duration of the post-stimulus undershoot without concomitant flow increase (~30s) clearly demonstrates a dissociation between microvascular response and brain oxygen consumption. We would like to point out that: interestingly, the time scale of the prolonged elevation in  matches well with previous observations in cultured neuronal tissues that a period of 30-40s is needed for restoration of ionic gradients (e.g. K+ and Ca2+) after neuronal activity is stopped (Brockhaus et al., 1993; Koch and Barish, 1994), suggesting that the reversal of ions across the membrane is the major component of brain energy use (Attwell and Laughlin, 2001) and the cause of the post-stimulus BOLD signal undershoot.

Despite the limited temporal resolution and signal to noise, it is of interest to look at the CBF, CBV, CMRO2, and BOLD changes at the start of activation. The data in Fig. 14d show that the CBV and CBF responses are preceded by the CMRO2 response but occur before the BOLD response. Interestingly, the BOLD response has not yet reached a maximum at 10 s after activation. This is similar to oxygen tension changes measured during somatosensory stimulation (Ances et al., 2001), the time dependence of which showed that the increase in tissue oxygenation following the initial dip was slower than the flow increase.

Therefore, multi-modal fMRI is useful in separating different metabolic and vascular responses and understanding the brain physiology.
Copyright 2006 Laboratory of FMRI, AIRC