Medial temporal lobe activation can be obtained in the elderly with an fMRI complex visual scene encoding paradigm. We found that activation in this brain region is greater in Normals than in patients with MCI and AD. The MCI patients were not significantly different from ADs. Our interpretation of this data is that the AD related neurodegenerative changes in the medial temporal lobe produce a functional (fMRI) impairment in MCIs that is of roughly equivalent magnitude to that observed in AD subjects. One measure of the potential sensitivity of any diagnostic test to early changes of AD is its ability to discriminate patients in the prodromal (MCI) phase from cognitively normal control subjects. Our results are encouraging in that regard. However, a more rigorous test of early diagnostic sensitivity is the ability of a test to predict whether or not individual elderly subjects will or will not eventually progress to AD. Our results do not address this issue at this time, but this is planned in the future.
We employed a passive sensory task as a control experiment to confirm that members of each clinical group on average had the potential to produce an equivalent BOLD response. There was no significant difference between the groups on the passive sensory task which we interpret to mean that the MCI and AD group’s depressed memory fMRI activation is not simply due to a globally impaired BOLD response. The results also imply that the depressed fMRI activation seen on the memory task in ADs and MCIs is not a non-specific disease related phenomenon, rather it is specific for the functional domain (memory) that clinically differs between cognitively normal and impaired subjects. When the memory and sensory results are considered together, they suggest that decreased temporal lobe activation may be a specific marker of medial temporal lobe dysfunction due to neurodegenerative disease, not a non-specific marker of old age.
Other groups have also assessed memory with fMRI in AD. Reduced fMRI activation has been reported in the hippocampal and entorhinal areas with memory tasks.6–9
Although we are not aware of any fMRI study that has specifically evaluated MCI subjects, our results agree in general with the aforementioned publications. Compensatory increased activation has been found in AD subjects during an encoding task in the medial parietal, posterior cingulate, and superior frontal regions.9
Using the ROC method, we were not in a position to evaluate all possible brain areas, and therefore cannot compare our results to studies in which extra temporal activation was evaluated.
Others have measured subregions of the medial temporal lobe (subiculum, entorhinal cortex) and found differences in memory activation among subgroups of individuals with memory decline relative to normals.8
The spatial resolution of the raw EPI images in our fMRI acquisition (3.75 mm × 3.75 mm × 5 mm) was coarser than the structures in question - the width of the entorhinal cortex or subiculum is in the range of three to four millimeters. We therefore elected to not attempt to isolate fMRI activation to specific subfields of the hippocampal formation (e.g., subiculum or CA1) or the entorhinal cortex.
There is no universally agreed upon method for analyzing fMRI data. The ROC method has advantages and disadvantages, and we do not mean to imply that it is necessarily the preferred method of fMRI analysis in all circumstances. An advantage is that it provides a systematic way to control for artifactual false positive “activation” in the image data across subjects. A disadvantage is that one must make a priori assumptions about where activation will occur. Unlike widely used brain mapping methods, the ROC method does not allow one to compare potential inter-group activation differences across the entire 3D volume of the brain. Despite this limitation, however, the ROC method may be quite useful for certain diagnostic questions, and it may also allow straightforward comparison of results between investigators. We employed well established functional-anatomic relationships for ROI definition in this study --medial temporal lobe for memory function, and para-central sulcus for primary sensory function. We felt that for this particular study, the improved reliability resulting from strict accounting of false positive artifactual activation was worth the associated trade-off of loss of anatomic coverage.
Another advantage of the ROC technique in general is the use of ROIs that are drawn on individual’s anatomy. Precise spatial registration among subjects is necessary if one is using the brain mapping approach to fMRI analysis - i.e., testing for inter-group differences in image space after spatial normalization. With the ROC approach, one doesn’t need to rely on precise spatial registration among subjects because the ROIs are drawn uniquely to each subject’s anatomy.
The particular implementation of the ROC method we used entailed calculating the true and false positive activation fraction in the same area of the brain for a given task. This has at least two useful properties. First, the number of pixels used to calculate the true positive and false positive fractions is identical. Second, different areas of the brain have their own unique noise properties. By using the same ROI for the true and false positive calculation, the noise properties are in theory better matched than if different areas of the brain were used to calculate the true vs. false positive fraction.
We acknowledge that there are inherent inaccuracies in tracing anatomic boundaries on near-isotropic T1-wighted images, and then superimposing these traces on spatially registered fMRI time series image data that were acquired as 5 mm thick axial slices. This is particularly true in the medial temporal lobe where susceptibility artifacts commonly cause anatomic distortion. Although we used a spin-echo echo planar acquisition specifically to minimize this phenomenon, some degree of distortion in susceptible areas is an inherent feature of most fMRI acquisition sequences. We drew the ROIs as “loose traces” around the medial temporal structures in order to account for the possibility of anatomic mismatch between the SPGR and the time series data. As a result we did not exclude areas of medial temporal activation from the ROI, but we did include some CSF in the medial temporal ROIs. Including CSF in the ROI decreased sensitivity by including inert–i.e., non-activating pixels -- in the calculation of TPF.
We limited the false positive fraction range of the ROC curves from 0 to 0.05. While this choice may be arbitrary, had we extended the false positive fraction range to greater than 0.05, we would have by definition been accepting greater than 5% false positive pixels as true positive activation. Limiting the false positive fraction range to less than 0.05 may be equivalent to the standard statistical practice of limiting the a priori probability of Type I error to less than 5%.
We did not directly assess responses during the memory activation paradigm, and therefore cannot guarantee that all subjects were fully attending to stimuli. We opted not to use handheld response units due to the difficulty subjects had in using the device as well as to reduce the probability of head motion during scans. Nonetheless, free recall on post scan testing corresponded to neurocognitive data and performance on the recognition test suggests that subjects were complying with directions.
Another limitation of this study, and fMRI in general, is the dropout rate due to motion corruption. Despite limiting the duration of each run as well as the entire scan time, we had to exclude 17% of our subjects because of excessive motion artifact. Advances in prospective motion correction engineering technology could substantially improve the reliability of fMRI for clinical purposes.
Our results indicate that the fMRI bold response of individuals with MCI during a memory paradigm is comparable to individuals with AD. Contrary to our expectation, the MCI and AD groups were indistinguishable in our data. Different activation paradigms and evaluation of activation in other brain regions with the objective of distinguishing MCI from AD may be a fruitful area for future studies. The results we have obtained while somewhat preliminary and focused only on medial temporal lobe activation, are nonetheless encouraging as they imply that fMRI is sufficiently sensitive to detect AD related changes in the prodromal, MCI, phase of the disease.