Moretti et al., EEG changes are specifically associated with atrophy in amydala and hippocampus in subjects with mild cognitive impairment (see: [140
Babiloni et al., Resting State Cortical Rhythms in Mild Cognitive Impairment and Alzheimer's Disease: Electroencephalographic Evidence [44
Deiber et al., Working memory electroencephalographic patterns in subtypes of amnestic mild cognitive impairment (see: [141
Olichney et al., Cognitive event-related potentials: Biomarkers of synaptic dysfunction across the stages of Alzheimer's Disease [142
Ashford et al., The topography of P300 energy loss in aging and Alzheimer's disease [143
Verdoorn et al., Evaluation and tracking of Alzheimer's disease severity using resting-state magnetoencephalograpy [144
Although studies of brain electrical activity have a long history in psychiatry and neurology, the advent of quantitative electroencephalography (qEEG) systems in the 1980 s introduced topographic mapping (“brain mapping”) as a display option. This important development brought EEG and related techniques squarely into the domain of neuroimaging. Onto a standardized head or brain template (or more recently onto the subject's own brain MRI) could be mapped the raw voltages of EEGs, averaged voltages of Evoked Potentials (EPs) and Event-related Potentials (ERPs), frequency domain measurements of EEG amplitude and power deriving from fast Fourier transformations (FFTs), results of inferential statistical tests such as significance probability mapping (SPM), and a wide range of other quantitative data. Simultaneously the technique of magnetoencephalography (MEG), recording magnetic instead of voltage fields produced by brain activity, made its debut, introducing magnetic counterparts to EEGs, EPs, and ERPs. Application of these new techniques to dementia in general and AD in particular was rapid. There are two broad paradigms for studying brain electrical activity. In one, the EEG/MEG eavesdrops on the resting or idling brain while the subject sits quietly with his eyes open or closed. Verdoorn et al., [144
] in this supplement present a vivid example of the use of the resting MEG to investigate AD. The other paradigm, subsuming EPs, ERPs and their magnetic equivalents, actively interrogates brain systems using external stimuli. In evoked potential (EP) studies auditory, visual, or other stimuli are used to drive the brain's sensory systems producing a sensory evoked potential containing a series of waves (components) corresponding to stages of cortical information processing. ERP studies elaborate on this framework by requiring the subject to perform a specific cognitive task related to the stimuli. The most common such task is the auditory oddball, in which the subject is instructed to ignore one class of stimuli (e.g., low pitch tones) but to respond to a second class of stimuli (e.g., high pitch tones). The brain responds with an ERP containing the familiar auditory sensory components followed by one or more new components (e.g., N200, P300) reflecting the additional information processing related to the cognitive task.
In many ways EEG offers an ideal method for assessing brain function. Its exquisite temporal resolution can track brain activity in the millisecond time domain characteristic of neuronal activity in the cortical substrate. It is entirely noninvasive and employs no ionizing radiation. It records both excitatory and inhibitory signals directly rather than secondary hemodynamic processes. It also is inexpensive. MEG offers these same advantages along with more precise spatial localization, although MEG systems are not in widespread clinical use due to their size and the necessity of super cooling their superconducting sensors with liquid helium. In contrast, EEG systems are abundant and in many cases portable.
Another important advantage of EEG is that normative databases are available, allowing statistical comparison of a patient's brain activity with that of age-matched controls. The use of quantitative techniques and inferential statistics moves EEG analysis from the realm of qualitative clinical impressions into the realm of quantitative empirical assessment. Such comparison with healthy controls yields information about the degree of abnormality of the patient's brain activity recorded by each electrode. Some databases additionally offer comparison with known clinical conditions, allowing a statistically based multivariate “best fit” classification that can aid clinical diagnosis. EEG's poor spatial resolution is being overcome by the use of increasingly dense electrode arrays, from 20 a decade ago to as many as 256 today. MEG, in addition to having a theoretically better spatial resolution than EEG, has experienced a similar increase in the number of sensors.
It has long been known that the typical EEG in AD contains increased slow activity in the theta (4–8 Hz) frequency range and decreased fast activity in the beta (13–24 Hz) range over the broad regions of the temporal and parietal lobes sustaining high levels of tissue damage from the disease [145
]. More localized cortical damage resulting from strokes produces more focal theta, and in principle it should be possible to use this to identify individuals suffering from vascular dementia [147
]. In practice however, this has been difficult to achieve using traditional univariate analysis techniques. Applications of multivariate techniques have shown more promise.
Quantitative EEG studies applying multivariate analysis to dementia have been reviewed extensively [148
]. Well-replicated studies have shown repeatedly that individual AD subjects and matched healthy controls can be classified into their appropriate groups on the basis of multivariate EEG analysis alone with accuracies as high as 80–90%. Furthermore, individual AD subjects could be discriminated from their nondemented depressed, alcoholic, or delirious, counterparts, and within the dementias AD subjects could be separated from those suffering from vascular or fronto-temporal dementia. However, such studies were performed using patients with established diagnoses and usually did not attempt to identify subjects in the earliest stages of a dementing process.
More recent work, reviewed in the Bablioni et al., [44
] and Moretti et al., (see: [140
]) articles in the Handbook and this supplement, greatly refines our understanding of the earliest frequency domain EEG changes in dementia. Subjects suffering from MCI were found to display several promising EEG markers. The markers not only distinguish between groups of MCI subjects and matched groups of healthy controls, but also between MCI sub-groups that will remain in MCI, progress to AD, or progress to non-AD dementia. It will be interesting to see whether these EEG markers, including integration with ApoE genotype, can be used to accurately classify individual subjects. If so, they could be employed as diagnostic aids and perhaps more importantly in a prognostic capacity. Additionally, the markers could serve as surrogate measures of disease progression, greatly aiding the development of new therapies.
The frequency domain changes seen in the EEG are paralleled by MEG changes. Verdoorn et al., [144
] in this supplement document MEG differences between groups of AD patients and healthy controls, and additionally find several MEG markers that change over time in parallel with neuropsychological changes to track disease progression. As with EEG, the critical question is whether MEG markers derived from groups of subjects can be applied to individuals. If so, they offer great potential for early phase development of novel treatments.
Pritchard et al., [150
] developed a new nonlinear mathematical method of analyzing EEG activity based on deterministic chaos theory, and derived a measure of brain activity they termed dimensional complexity. They then used dimensional complexity to study AD and found that not only did this measure reliably distinguish between groups of AD patients and groups of matched healthy controls [151
], but it also could reliably classify individuals as belonging to either of these two groups [146
]. Direct comparison between standard frequency analysis and a combination of frequency analysis and dimensional complexity clearly showed the superiority of the combined technique. The use of nonlinear dynamic analysis has been limited by the availability of computational power. Indeed, those early studies required collaboration with the Supercomputer Computations Research Institute at Florida State University. But in the two decades since those seminal studies, rapid increases in computational power have allowed the analyses to be run on desktop computers, and nonlinear analysis has occupied a minor but important role in EEG research. Bablioni et al., [44
], in this supplement, reviews some recent nonlinear dynamic findings regarding AD (e.g., the sparing of resting state posterior alpha EEG rhythms in AD patients with more severe ischemic changes in the white-matter).
Because AD involves widespread brain pathology and marked deterioration of cognitive functions one might expect changes in both EPs and ERPs, and both are seen. For example, the visual EP in response to a diffuse light flash contains a P2 component that has been found consistently to be delayed in groups of AD patients [145
]. This delay probably reflects damage to the cholinergic neurons in visual association areas of the cortex. Similarly, the ERP produced by AD victims during the oddball task typically contains a delayed P300 component, probably reflecting the additional processing time necessary for the damaged higher-order association areas of the cortex to perform the cognitive task. The amplitude of the P300 component is often found to be diminished in AD, presumably reflecting a reduced population of cortical pyramidal neurons involved in the cognitive oddball task. Unfortunately, neither the latency increase nor the amplitude decrease is sufficiently reliable to be of clinical value when assessing individual patients. In an effort to extract a more reliable P300 signal from the background noise, Ashford et al., [143
], in this supplement, compute power and energy measures from the recorded P300 voltage record. Both derived measures appear to track age- and AD-related changes more closely than does the traditional voltage wave.
In this supplement, Olichney et al., [142
] review prior ERP studies of AD, including P300 studies of attention and N400 studies of linguistic processing. Importantly, ERP studies can be designed to be sensitive to the cardinal features of AD. In this regard, recent work by Olichney and colleagues suggests that a Late Positive Component important for memory processes, sometimes termed the P600, may be particularly sensitive to the earliest stage of synaptic dysfunction during the ‘Pre-clinical’ (MCI) stage of AD. Olichney et al., [157
] have demonstrated that two late ERP components (N400 and P600) are also promising in their ability to predict outcome in MCI. As with the EEG markers proposed by Bablioni et al., [44
], an important question is whether ERP markers can accurately classify individual subjects during the MCI stage or even earlier. Recently proposed research criteria for pre-clinical AD [20
] divided this entity into 3 stages based on the presence/absence of very mild cognitive deficits and synaptic dysfunction.
This supplement illustrates several applications of the EEG, ERP and MEG techniques to characterize synaptic/neuronal function and their earliest derailments in AD. Further research and validation of these measures are needed to test their clinical utility and cost-effectiveness and to determine which information is most complimentary to the results from other imaging modalities (e.g., MRI, PET) and other AD biomarkers.