Fig. 1
Incidence of known types of dementias
The AD is the most common type of dementia accounting for an estimated 60–80 % of all cases. In the AD, information transfer at the synapses begins to fail, the number of synapses gradually declines and eventually brain cells die [4–6]. According to International Classification of Diseases—ICD-10 [7], the AD is classified as mental and behavioral disorder. The disease took its name by the German neurologist Alois Alzheimer, who first described the symptoms and the neuropathology of the disease in 1906. He identified the two major abnormalities in the brain that characterize the disease—senile plaques and neurofibrillary tangles—localized to specific regions of the brain which control memory and learning [3–5].
The AD is a gradually progressive disorder with a course of 6–12 years. It is diagnosed in people over 65 years of age, although the early-onset of the disease can occur much earlier. There is evidence that brain abnormalities are present for at least 30 years before symptoms become apparent [3, 8]. Although AD develops differently for every individual, memory loss, difficulty performing familiar tasks, problems with language, disorientation to time and place, poor or decreased judgment, problems with abstract thinking, misplacing things, changes in mood and behavior, changes in personality, and loss of initiative are the ten most common warning signs of the disease [3–6].
The development of AD seems to correlate well with increasing age; 10 % of people over the age of 65 and 50 % over 85 suffer from AD. While scientists know that AD involves failure of nerve cells, they do not know why this happens. They agree that the disease is not only a result of a single cause but a combination of multiple factors, with race, occupation, level of education, sex, geographic location, lifestyle, socioeconomic status contributing to the onset of the disease but are not considered decisive risk factors. Other important factors for the disease include family history, brain trauma, metabolic disorders, regulation of amyloid precursor protein and tau protein, as well as, the genetic factors [3–6, 9–13].
Current consensus statements have emphasized the need for early diagnosis of AD [13]. However, there is no single, comprehensive method to diagnose AD. Based on medical history, physical examination, cognitive assessment tests, laboratory and brain imaging results, a properly trained physician with expertise in dementia may diagnose AD. The diagnosis is made using the National Institute of Neurological and Communicative Disorders and Stroke—Alzheimer’s Disease and Related Disorders Association criteria (NINCDS-ADRDA) [14–16] and the criteria of Diagnostic and Statistical Manual of Mental Disorders (DSM-IV) of the American Psychiatric Association [17]. Typically, it takes a few weeks to complete a diagnostic evaluation of the disease and most often diagnosis is achieved at late stages of the disease. Thus, it is important to support the clinician with tools and methods that facilitate more effective diagnosis at earlier stages of the disease. Some of these tools relate to anatomical/structural and functional characteristics of the brain that have been associated with and degraded by the disease.
Imaging assessment plays an important role in the diagnostic workup of patients with AD. A variety of neuroimaging techniques, including positron emission tomography (PET), functional magnetic resonance imaging (fMRI), single-photon emission computed tomography (SPECT) and structural magnetic resonance imaging (MRI) have made significant advances over the last years towards the early diagnosis of the disease and the monitoring of its progression [13, 18–21].
Imaging modalities allow clinicians and researchers to better examine, measure, and understand how different cognitive functions are affected by the presence of AD. Such methods allow for the assessment of “where” activity is localized in the brain. In an attempt to understand “when” such activity happens—how brain response changes over time—typical modalities measure the electrical fields generated by the active neural populations [22]. For this purpose, electroencephalogram (EEG) has rapidly gained growing interest and it is used more and more in clinical practice as a method for the assessment of AD. Abnormalities of cortical activity, due to AD (cortical dementia), are depicted as abnormalities in EEG, since the electrical signals recorded in EEG are scalp potentials reflecting underlying cortical activity.
The aim of this chapter is to broadly examine and critically review research utilizing fMRI and EEG techniques separately to better understand AD and improve the effectiveness of medical diagnosis of the disease. Attention is given to studies that are based on the EEG-fMRI information fusion, which exploit the complementary nature of the two modalities in order to support diagnosis of AD. We should notice here that in this chapter we focus at the functional level of brain operation as affected by the disease. In a holistic consideration of the disease and its diagnosis, this functionality should be combined with information at cellular and genomic level, as well as with anatomic effects (e.g., size of white matter) and with regular cognitive tests, but this is out of the scope of this chapter.
2 Assessment of Alzheimer’s Disease Based on Functional Magnetic Resonance Imaging
Functional Magnetic Resonance Imaging (fMRI) is a noninvasive neuroimaging technique which uses the physical phenomenon of Nuclear Magnetic Resonance (NMR) and the associated technology of MRI to measure and localize specific human brain functions. It works by detecting regional changes in cerebral metabolism, in blood flow, volume, or oxygenation in response to task activation. It relies on the fact that cerebral blood flow and neuronal activation are coupled. When a brain area is more active, it consumes more oxygen and to meet this increase demand blood flow increases to the active area [23–25].
The fMRI signal depends on the vascular response to functional brain activation and it is typically implemented by imaging of the Blood Oxygen Level Dependent (BOLD) contrast, which is based on the differentiation of the magnetic properties of oxygenated (diamagnetic) and deoxygenated (paramagnetic) blood. These magnetic susceptibility differences lead to small but detectable changes in weighed MR image intensity. More specifically, when the human brain receives a stimulus an increase in neuronal activation takes place. Since neurons do not have internal reserves of energy their activation causes a need for more energy. The energy is supplied in the form of glucose and oxygen, which is carried in hemoglobin. The resulting increased need for oxygen is over-compensated by a large increase in perfusion. As a result, the venous oxyhemoglobin concentration increases and the deoxyhemoglobin concentration decreases. The presence of deoxyhemoglobin causes local field inhomogeneities, which are responsible for a dephasing of the local transversal magnetization, leading to a reduction in the transverse relaxation time T2. As a diamagnetic molecule oxyhemoglobin does not produce the same dephasing. Thus, changes in deoxyhemoglobin can be observed as the BOLD contrast in T2-weighted magnetic resonance images, serving as an indirect measure of neuronal activity [23–26].
Among the several distinct advantages of the fMRI image modality (it is considerable safer since no contrast agent is needed in order the signal to be administrated, the total scan time is very short, it has high spatial and temporal resolution) is that its application in clinical practice allows neuroscientists to investigate the relationship between brain and behavior. It acts as a tool for identifying brain regions that are associated with certain perceptual, cognitive, emotional, and behavioral functions. It is suitable for accessing many aspects of human cognition and plays an important role in assisting pre-surgical planning in neurosurgery and in providing additional diagnostic information in the clinical environments of patients who have functional disorders due to neurological disease and mental illness. Some other fundamental application of fMRI is the management of pain, the improved assessment of risk, and the improved seizure localization [23, 25].
In this section we focus on how fMRI can contribute and improve the assessment of AD. A large number of studies have been reported in the literature regarding the analysis of fMRI in patient with AD. The studies can be grouped into two broad categories. The first category includes studies concentrated on identifying what is differentiated in brain function between healthy subjects and subjects with AD. The second category includes studies that are trying to express the above-mentioned differences through an index that can serve as a biomarker for the diagnosis of AD.
2.1 Functional Magnetic Resonance Imaging Analysis Studies Revealing Differentiations Between Healthy and Alzheimer’s Disease Subjects
Clinical studies have shown that the posterior cingulate cortex (PCC) presents reduced activity in patient with AD, as well as, hypo metabolism in cognitively intact subjects with genetic susceptibility to the disease [27–32]. The combination of clinical findings led Raichle et al. [33] to propose that the PCC forms part of a “default mode” network, a network of brain regions that are active when the individual does not perform any task and the brain is at wakeful rest. Greicius et al. [34] using functional connectivity analysis of fMRI data observed that there is significant coactivation of several regions within this network.
Rombtous et al. [35] studied the activity and reactivity of the default mode network in the brain in patients with mild cognitive impairment (MCI), with mild AD and in healthy controls (HC). The three groups are differentiated in the early phase of deactivation. More specifically, the response of the default mode network in anterior frontal cortex distinguished MCI from HC and AD, while the response in the precuneus could only distinguish between patients and HC. These differences reflect the reactivity and adaptation of the network. For the conduction of the study they utilized fMRI analysis methods included in software packages like FEAT and FSL.1
The study of Sorg et al. [36] revealed the reduction or absence of functional connectivity and the atrophy of the regions consisting resting state networks in patients with AD. More specifically, Independent Component Analysis (ICA) of resting-state fMRI data was utilized in order eight spatially consistent resting state networks to be identified. Only selected areas of the networks demonstrated reduced activity in the patient group. Atrophy was presented in both medial temporal lobes of the patients through voxel-based morphometry. Regarding functional connectivity between medial temporal lobes and the posterior cingulate of the network it was present in healthy controls but not in patients.
Oghabian et al. [37] focused their study on the evaluation of the fMRI in differentiating between Alzheimer’s, MCI and healthy aging. They applied ICA to compare the resting-state brain activation patterns between groups of subjects examining in this way the ability of fMRI to differentiate the three conditions mentioned above. The Minimum Description Length algorithm was employed to determine the number of independent components [38]. The components were transformed to Z space and then a Gaussian Mixture Model was applied to define the value of the threshold determining the creation of the activation maps [39]. Their studies revealed that healthy aging brain presents activation areas with larger area and greater intensity of activation compared with the MCI and Alzheimer’s group in PCC region of the brain. It must be mentioned that the observation is valid when the subjects are under a certain resting-state session.
Small et al. [40, 41] studied how the hippocampal regions are interconnected to form a circuit. Using fMRI they evaluated the hippocampal regions in vivo, and they used their method to study three different groups of subjects, elderly with normal memory, elderly with isolated memory decline, and elderly with probable AD. Two distinct patterns of regional dysfunction were revealed among the last two groups. The first pattern involved all hippocampal regions, while the second one present dysfunction restricted to the subiculum. The processing of fMRI data was achieved using appropriate software packages (MEDx Sensor Systems, Boulder, CO; IDL Research Systems, Sterling, VA).
It has been observed that the development of AD causes neuropathological changes which are followed by the decline of basic cognitive processes. The two questions that arise are the following: (a) Does the functional alterations in neural circuitry accompany these neuropathological changes? and if so (b) Can they be detected before onset of symptoms?. Based on these observations Smith et al. [42] used fMRI to examine if and how cortical activation is differentiated between cognitively normal subjects and subjects at risk for developing AD. The analysis of the data revealed that the two groups present similar patterns of brain activation, while the high risk group showed areas of significantly reduced activation in the mid- and posterior inferotemporal regions. AFNI2 software and more specifically, cubic interpolation with splines, was utilized for the resampling of the activation maps and appropriate MATLAB 5 (MathWorks, Natick, MA) functions for the smoothing of the maps. The results of the previous processes were compared voxel by voxel using a two sample t-test.
Machulda et al. [43] based on the fact that the main feature of patients with AD is memory impairment studied whether an fMRI memory encoding task can distinguish cognitively healthy elderly individuals, patients with MCI, and patients with early AD. The following conclusions have been obtained: (a) MCIsubjects and subjects with AD had less medial temporal lobe activation on the memory task than the healthy subjects, but similar activation as healthy subjects on the sensory task, (b) decreased medial temporal activation may be a specific marker of limbic dysfunction due to the neurodegenerative changes ofAD, (c) fMRI can be used to detect changes in the prodromal, MCI, phase of the disease. The fMRI data were processed using AFNI software.
Petrella et al. [44] exploited the potentiality of the fMRI to examine how memory networks break down as AD progresses. They tried to detect brain regions in which changes in activation correlate with degree of memory impairment across Alzheimer disease, MCI, and elderly control subjects. They utilized a face-name associative encoding paradigm and the results revealed preservation of some areas of activation with an overall decrease in activation of medial temporal lobe structures from control subject to patient with MCI to patient with mild AD. Increase of activation magnitude was noted in the posteromedial cortex (PMC) region. PMC activation changes were larger in both magnitude and extent than those in the MTL. In addition, changes in PMC regions are significantly correlated with neuropsychological test performance.
Gron et al. [45] assessed episodic memory in older subjects that were diagnosed with probable Alzheimer’s or major depressive disorder. In the fMRI paradigm repetitive learning and free recall of abstract geometric patterns were used. The analysis of fMRI data was performed using Statistical Parametric Mapping (SPM99, Welcome Department of Cognitive Neurology, London, UK) and revealed that: (a) healthy seniors or depressive patients present superior hippocampal activation than patients with AD, (b) patients with AD showed bilateral prefrontal activity while healthy seniors did not, (c) patients who had major depressive disorder, in contrast to healthy seniors or patients who had AD, showed activation of the orbitofrontal cortex and the anterior cingulate in the hemispheres of the brain. In similar conclusions, regarding the differentiation of activation patterns of memory correlated regions, were presented by Sperling et al. [46, 47].
Differences between patients with probable AD (pAD) and cognitively able elderly volunteers, based on a visuospatial cognition task, were studied by Thulborn et al. [48]. They analyzed fMRI data received during an eye movement paradigm. The study concluded that: (a) statistically significant differences existed between the activation patterns of the patients with pAD and those of the volunteers, and (b) a left-dominant parietal activation pattern and an enhanced prefrontal cortical activation were observed in most patients with pAD but not in the control group. A three-phase analysis was followed. First the artefcats were removed and then the activation maps were created using a voxel-wise t-test. Finally, the AFNI software was used to detect the regions with higher activation. Talairach coordinates, activation volumes, and laterality ratios were used to characterize the activation patterns.
Buckner et al. [49] examined if the properties (amplitude and variance)) of hemodynamic response are differentiated between non demented and demented older adults. They conducted an fMRI event-related design paradigm involving repeated presentation of sensory-motor response trials. The results showed that visual cortex present significant reduction in the amplitude of the hemodynamic response function in contrast to the variance of the hemodynamic response that was significantly increased in demented than in non demented older adults. In the motor cortex the characteristics of the hemodynamic response were not affected in both groups of subjects. For the extraction of the results the fMRI data were preprocessed using SPM99 and ANOVA software.
Prulovic et al. [50] examined functional activation patterns in patients with AD during active visuospatial processing. They examined also how the local cerebral atrophy affects the strength of the activation. The fMRI data were recorded while subjects performed an angle discrimination task. Patients with AD and healthy subjects presented overlapping networks (superior parietal lobule, frontal and occipito-temporal cortical regions, primary visual cortex, basal ganglia, and thalamus). Superior parietal lobule and occipito-temporal cortical were the regions of the network that presented the most pronounced differences between the two groups of subjects. The differences can be attributed to the differences in individual atrophy of superior parietal lobule region. They suggested that the local cerebral atrophy should be considered as an expected result of all functional imaging studies of neurodegenerative disorders. The fMRI data were analyzed using the Brain Voyager Software.3
The conclusions presented in the study of Hao et al. [51] were based on the processing of fMRI data produced during two types of visual search tasks. The detection of the anatomical areas of activation that are associated with visual attention processing and the detection of changes that may occur in a group of AD patients compared with a control group was the purpose of the study. The findings of the study suggested that patients with AD present reduced activation in both parietal lobes and the left frontal regions. On the other hand, the healthy subjects presented increased activation in the right frontal lobes and the right occipito-temporal cortical regions with the conjunction task.
The relationship between fMRI activation and measures of global and regionally specific atrophy in normal aging and in Alzheimer disease was studied in [52]. The study was conducted by Jonshon et al. [52] and revealed that during a semantic process the left inferior frontal and the left superior temporal gyri were activated. Thus, the evaluation of the correlations between those regions and measures of local atrophy followed. No significant correlation exist in healthy subjects, in contrast to AD patients that present positive and negative correlation between atrophy and activation in left inferior frontal and left superior temporal gyri, respectively.
Rombouts et al. [53] analyzed fMRI data from healthy controls and patients with AD in order to study brain activation during a learning task. The aim of the study was to test the hypothesis that brain activation is decreased in the medial temporal lobe memory system in patients with AD compared to controls. The study revealed that the fusiform, parietal and occipital parts, the hippocampal formation and the frontal cortex present activation. Regarding the medial temporal lobe the results were differentiated between the two encoding tasks. In the first task a significant bilateral decrease of the activation in the left hippocampus and parahippocampal gyrus was observed in patients with AD, compared to control volunteers. Nothing of the above was observed the second encoding task. The analysis was conducted with the SPM99 software. Next, the data regarding each participant were modeled using a box car design, convolved with the hemodynamic response function. Then for each participant a “contrast image,” was calculated. Each participant’s contrast-enhanced image was then fed into a statistical test. A one-sample t-test was used for the assessment of the average brain activation and a two-sample t-test for the assessment of the differences in brain activation between patients and control volunteers.
The effect of AD on functional neuroanatomical processing of semantic and phonological information was studied by Saykin et al. [54]. Healthy controls and AD patients present predominant activation foci in the inferior and middle frontal gyrus. However, patients present additional activation in the left prefrontal cortex.
The relationship between brain responses to memory tasks and the genetic risk factor (epsilon4 allele of the apolipoprotein E gene—APOE) is examined in the study of Bookheimer et al. [55]. AD affects both the magnitude and the extent of brain activation. More specifically, subjects with the APOE epsilon4 allele present larger brain activation, in terms of magnitude and extent, than subjects with the APOE epsilon3 allele. The signal intensity presented a larger, on average, increase during periods of recall in the carriers of the APOE epsilon4 allele. In addition, they presented a higher mean number of activated regions throughout the brain than the carriers of the APOE epsilon3 allele.
Important differences between healthy and demented subjects, as far as it concerns the anatomic distribution of the activation especially in the regions of posterolateral temporal and inferior frontal cortex, were presented in the study of Grossman et al. [56]. The patterns of neural activation were created by the analysis of fMRI data, using SPM99, which is formed during a verb processing task.
The studies reported until now examined the differentiation in brain regions which show positive activations, defined as increases in signal during an active task compared with a more passive baseline. Lustig et al. [57] focused on examining how negative activations, or deactivations, defined as decreases in signal during the task as compared with baseline, are differentiated between healthy (young and older adults) and demented subjects. Their study showed that deactivation in medial frontal regions was reduced due to aging but was not affected by the AD. On the other hand, the medial parietal/posterior cingulated region presents significant differences between healthy and demented subjects. Furthermore, the temporal profile of the region response was initially activated by all groups, but the response in young adults quickly reversed sign, whereas AD individuals maintained activation throughout the task block.
Celone et al. [58] studied the positive and negative activations that take place in memory related regions. The independent component analyses, performed on fMRI data, revealed specific memory-related networks that activated or deactivated during an associative memory paradigm. During the course of MCI and AD there is a direct relation between the loss of functional integrity of the hippocampal systems, responsible for memory functions, and the alterations of neural activity in parietal regions. These data may also provide functional evidence of the interaction the between neocortical and the medial temporal lobe pathology in early AD.
The studies reported above reveal significant differences between healthy and AD patients. The differences are related to:
(a)
the activation patterns of specifically selected brain regions such as memory related regions, regions of the visual and motor cortex as well as regions responsible for the learning processes,
(b)
the functional connectivity of the default mode network,
(c)
the task-induced deactivation of brain networks, and
(d)
the regional homogeneity of the fMRI signal.
A detailed review can be found in [59, 60]. For the detection of the differences fMRI data, that was received during resting-state or during task requiring the stimulation of cognitive functions, were preprocessed and analyzed through the utilization of statistical and connectivity analysis methods [26]. The studies revealed the potentiality of fMRI as a tool for the assessment of AD in contrast to the methods which are described in the next section that quantify differences detected using fMRI data and propose an index for the diagnosis of AD.
2.2 Functional Magnetic Resonance Imaging Analysis Studies Proposing Markers for the Diagnosis of Alzheimer’s Disease
As already mentioned above a large number of studies is concentrated in the alternations of functional connectivity and activation patterns of brain networks that are activated even though the subjects are in a resting state and they do not perform a cognitive demanding task. Greicius et al. [61] examined the functionality of such a network and revealed that all group of subjects (healthy adults and AD patients) presented coactivation in the region of hippocampus indicating that the default-mode network is closely involved with episodic memory processing, the disrupted connectivity between posterior cingulate and hippocampus is responsible for the posterior cingulate hypometabolism and the activity in the default-mode network may be a biomarker of AD with 85 % sensitivity and 77 % specificity. The results of their study were based on the analysis of fMRI data. More specifically, they employed independent component analysis using FSL MELODICA ICA,4 and the four best components were selected. The selected components were combined using random effect analysis methods to create the activations maps for each group of subjects. A two sample t-test was applied for the comparison of the maps of each group. For further comparison between groups a goodness-of-fit analysis was applied.
The study of Supekar et al. [62] focused on the same network. The network was created for 18 healthy and 12 demented subjects and it was studied as an undirected graph. Wavelet analysis was applied for the computation of frequency-dependent correlation matrices that were thresholded in order the undirected graph of the functional brain network to be created. The characteristic path length and the clustering coefficient were computed using graph analytical methods. The analysis of those coefficients revealed significant differences between the two groups, differences capable to distinguish AD participants from the controls with a sensitivity 72 % and specificity 78 %.
The study of Wang et al. [63] was based in the regions of one task-positive and one task-negative network. The networks consisted of five bilateral homologous regions and 6 bilateral homologous regions, respectively. The mean time series of each of the regions were extracted by averaging the fMRI time series over all voxels in the region. Correlation coefficients were computed between each pair of the regions and a Fisher’s r-to-z transformation was applied to improve the normality of these coefficients. The correlation coefficients between each pair of regions were entered into an one-sample two-tailed t-test to determine the existence or not statistical significant differences between the coefficients. Then Pseudo-Fisher Linear Discriminative Analysis (pFLDA) was performed to generate a linear classifier. The application of the above procedure to 14 health and 14 demented subjects showed a correct classification rate 72 % for the healthy subjects and 93 % for the demented.
Zhang et al. [64] studied alterations in functional connectivity in the resting brain networks in healthy elderly volunteers and patients with mild, moderate, or severe Alzheimer Disease. They preprocessed the fMRI data to remove artifacts and they segmented the data in gray matter and cerebrospinal fluid. The segmented gray matter data were filtered using a phase-insensitive band-pass filter in order the effect of low-frequency drift and high-frequency physiologic noise to be reduced. Posterior cingulated cortex was selected as the region of interest and Pearson Linear correlation coefficients between the time series of each scaled voxel and the time series of the average signal of the posterior cingulated cortex were computed. A Fisher z transform was applied to improve the normality of these correlation coefficients. The whole analysis revealed that the functional connectivity of the default mode network is affected by the presence and progression of AD.
Li et al. [65] tried to express the differentiations presented between nine healthy subjects, five subjects with mild cognitive impairment and ten subjects with AD using cross-correlation coefficients of spontaneous low frequency—COSLOF index, which is defined as the mean of the cross-correlation coefficients of spontaneous low frequency components between possible pairs of voxel time courses in a brain region. The two-tailed Student t-test was used to determine differences in the COSLOF index between the three groups. The results showed that the COSLOF index can differentiate the three groups with 80 % sensitivity and 90 % specificity.
Based on the previous work, Xu et al. [66] studied the effect of signal to noise ratio and the phase shift of spontaneous low-frequency (SLF) components on the index. They propose an new index called phase shift index—PHI. The application of the PHI on three groups of subjects (the control, mild cognitive impairment and AD) demonstrated that PHI is high in patients with AD (72.6 ± 11.3) in contrast to patients with mild cognitive impairment (58.6 ± 5.7) and healthy subjects (40.6 ± 8.4). They claimed that the PHI is more reliable index than the COSLOF index, however, they do not provide evaluation measures to support this fact.
Chen et al. [67] and Burge et al. [68] employed the properties of a Bayesian network classifier to diagnose AD. Chen et al. [67] utilized a Bayesian network classifier with inverse-tree structure to detect the brain regions with activation maps that can lead to the diagnosis of the disease. Burge et al. [68] constructed a dynamic discrete Bayesian network classifier which recognizes the functional correlations between neuroanatomical regions of interest and examines if those correlations can distinguish healthy from demented subjects. The above-mentioned studies diagnosed AD with 81 and 70 % accuracy, respectively.
Tripoliti et al. [69] proposed a six-step method (Fig. 2). First, spatial and temporal preprocessing of fMRI data were conducted to remove artifacts and the subsequent analysis to lead to reliable results (Step 1). A generalized linear model was applied to the voxel time courses extracted from the preprocessed images (Step 2). At the end of the second step an activation map was created for each subject. The analysis of each map allowed the extraction of features describing the activation patterns and the hemodynamic response of the brain. An index (path length), expressing the head movement during the conduction of the fMRI experiment, was calculated from the functional images. Additionally, the anatomic MRI images were exploited to extract features describing the atrophy of gray matter (Step 3). The extracted features, as well as, features recorded during the experiment (demographic details—age and sex and behavioral data—median and average reaction time) were evaluated for their potential to diagnose AD and monitor its progression. The subset of features, selected by the application of an information theory based selection approach (Step 4) was supplied as input to the classification step (Step 5). The Random Forests algorithm and Support Vector Machines (SVMs) were evaluated for their classification accuracy, sensitivity and specificity. Finally, the trees that consist of the forest were converted to rules (Step 6). The rules combine the information, which is encompassed to the features, and they provide the decision if the subject is suffering from AD. In the case of positive diagnosis the rules determine the stage of progression (very mild or mild). The method is evaluated using a dataset of 41 subjects. The results of the proposed method indicate the validity of the method in the diagnosis (accuracy 94 %) and the monitoring of the AD (accuracy 97 and 99 % for the three and four class problem, respectively). By modifying the Random Forests algorithm the disease was diagnosed with 98 % accuracy [70].
A short review of the above-mentioned studies can be found in Table 1. In the next section we focus on the potential of EEG to provide biomarkers for the assessment of AD.
Table 1
A short review of studies regarding assessment of Alzheimer disease based on fMRI
Authors | Subjects | Design | fMRI analysis methods | Main conclusions | |
---|---|---|---|---|---|
1. | Small et al. [40] | 4 healthy12 with MCI4 with AD | Memory related task | Pixel by pixel t-test | Memory impairment is associated with the dysfunction of the hippocampus |
2. | Saykin et al. [54] | 6 healthy9 with AD | Auditory stimulation tasks | Statistical analysis—SPM96 | Patients presented additional activation in left prefrontal cortex |
3. | Smith et al. [42] | 12 low risk14 high risk | Visual naming and letter fluency tasks | Cubic interpolation with splines, appropriate MATALAB 5 functions, two sample t-test | The high risk group showed areas of significantly reduced activation in the mid- and posterior inferotemporal regions |
4. | Bookheimer et al. [55] | 14 no carriers of ApoE416 carriers of ApoE4 | Memory related paradigm | SPM96—ROI analysis | The carriers of the APOE epsilon 4 allele presented greater activation in terms of magnitude and extent than the carriers of the APOE epsilon3 allele. The signal intensity presented a greater, on average, increase during periods of recall in the carriers of the APOE epsilon4 allele. They presented a greater mean number of activated regions throughout the brain than the carriers of the APOE epsilon3 allele |
5. | Buckner et al. [49] | 14 healthy young14 healthy old13 with AD | Sensory-motor task | Hemodynamic Response Function Analysis | Motor cortex: no differencesVisual cortex: significant decrease was presented in the amplitude of the hemodynamic response function in AD patients |
6. | Johnson et al. [52] | 16 healthy8 with AD | Visual search tasks | SPM96 | AD patients presented positive and negative correlation between atrophy and activation in left inferior frontal and left superior temporal gyri, respectively |
7. | Small et al. [41] | 4 healthy4 with AD | Memory related task | Independent t-test ROI analysis | The patients with AD presented reduced activation in the hippocampus and the subiculum |
8. | Thulborn et al. [48] | 10 healthy18 with AD | Visually guided saccade paradigm | Voxel wise t-test | The patients with pAD present significant differences in the activation patterns (a left-dominant parietal activation pattern and enhanced prefrontal cortical activation) compared to the volunteers |
9. | Rombouts et al. [53] | 10 healthy12 with AD | Two learning tasks | SPM99, one sample t-test and two sample t-test | The fusiform, parietal and occipital parts, the hippocampal formation and the frontal cortex presented activationAD patients vs. healthy controls: decrease activation in the left hippocampus and parahippocampal gyrus during the first encoding task but not during the second |
10. | Prvulovic et al. [50] | 14 healthy14 with AD | Angle discrimination task | Brain Voyager software | Superior parietal lobule and occipito-temporal cortical were the regions of the network that presented significant differences between the two groups of subjects |
11. | Gron et al. [45] | 12 healthy12 with MCI12 with AD | Repetitive learning and free recall of abstract geometric patterns | Statistical analysis—SPM99 | Memory presents significant dysfunction in patients with Alzheimer’s disease in contrast to healthy subjects and to patients with memory decline due to major depressive disorders |
12. | Li et al. [65] | 9 healthy5 with MCI10 with AD | Resting-state | ROI analysisCOSLOF index | COSLOF index is significant lower in patient with Alzheimer’s disease compared with health subjects and subjects with MCI |
13. | Grossman et al. [56] | 16 healthy11 with AD | Subjects judged the pleasantness of verbs, including MOTION verbs and COGNITION verbs | SPM99 | The anatomic distribution of the activation especially in the regions of posterolateral temporal and inferior frontal cortex, presented significant differences |
14. | Sperling et al. [46] | 10 healthy young10 healthy old7 with AD | Face–name association encoding task | Statistical analysis—SPM99 | Healthy old adults vs. younger controls: (a) reduced activation in both superior and inferior prefrontal cortices, (b) greater activation in parietal regionsAlzheimer patients: (a) reduce activation in the hippocampal formation, (b) greater activation in the medial parietal and posterior cingulate regions |
15.
![]() Stay updated, free articles. Join our Telegram channel![]() Full access? Get Clinical Tree![]() ![]() ![]() |