Advertisement

Diagnosis of Alzheimer’s Disease Using Brain Imaging: State of the Art

  • Atif ShahEmail author
  • Kamal Niaz
  • Moataz Ahmed
  • Reem Bunyan
Chapter
  • 519 Downloads

Abstract

Alzheimer’s disease (AD) is one of the prominent diseases in elderly people which leads to language impairment, disorientation, memory loss, and eventually death. Despite the severity of the disease, there is no such drug reported to control, reduce, or stop the progression of AD. The neuroimages played a crucial role in tracking the progression of AD using biomarkers which help the physicians to diagnose the disease more accurately. In this study, we investigate the effectiveness of structural and functional neuroimaging modularities which are used in the state-of-the-art methods to diagnose AD. The finding shows that most of the studies prioritize magnetic resonance imaging techniques (MRIT) solely or combined with other neuroimaging modularities to achieve better performance. Studies also founded that only few public datasets are available, and the most widely used public dataset is Alzheimer’s Disease Neuroimaging Initiative.

Keywords

Medical imaging MRI Alzheimer’s disease CAD Features extraction 

Abbreviations

3D

Three dimensions

ACC

Accuracy

AD

Alzheimer’s disease

ADNI

Alzheimer’s Disease Neuroimaging Initiative

AUC

Area under the curve

Amyloid-β

CAD

Computer-aided diagnosis

CNN

Convolutional neural network

CR

Creatine

CSF

Cerebrospinal fluid

CT

Computed tomography

DL

Deep learning

DTI

Diffusion tensor imaging

DWI

Diffusion-weighted imaging

FC

Functional connectivity

FDG

Fluorodeoxyglucose

fMRI

Functional magnetic resonance imaging

FN

False-negative

FP

False-positive

GP-LR

Gaussian process logistic regression

ICA

Independent component analysis

LLE

Local linear embedding

MCI

Mild cognitive impairment

MCI-A

Amnestic MCI

MCI-C

MCI converted

MCI-NC

MCI non-converted

MCI-P

Progressive MCI

MCI-S

Stable MCI

MRI

Magnetic resonance imaging

MRS

Magnetic resonance spectroscopy

NAA

N-Acetyl aspartate

PCA

Principal component analysis

PET

Positron emission tomography

RBF

Radial basis function

re-fMRI

Resting-state fMRI

RFE

Recursive feature elimination

ROC

Receiver operating characteristic

ROS

Reactive oxygen species

SEN

Sensitivity

SPE

Specificity

SPECT

Single photon emission computer tomography

SVM

Support vector machine

T1-w

T1-weighted

T2-w

T2-weighted

TN

True-negative

TP

True-positive

VBM

Voxel-based measure

WM

White matter

WMH

White matter hyperintensities

7.1 Introduction

Alzheimer’s disease (AD) is the utmost common cause of cognitive impairment usually in elderly people; around 11% of people at the age of 65 and above suffered from AD (Alzheimer’s Association 2014). AD progression leads to memory loss, disorientation, and language impairment and eventually leads to death, which increased to 123% between 2000 and 2015 (Alzheimer’s disease facts and figures 2018). It is well understood that the prevalence and severity of AD are high, still there is no drug or treatment that has been reported to reduce the risk or completely stop AD progression. The impact of AD can be understood via the progressive brain changes associated with AD. To achieve this, brain imaging played a profound role in track changes in the follow-up scans using different imaging techniques and transformed AD research and clinical practices. Brain imaging provides informative biomarker even before clinical symptoms have appeared. These biomarkers help in the diagnostic decision and disease assessment and improve diagnosis. The use of a biomarker for AD will also be helpful for clinical trials because of its enrichment strategies or tough inclusion criteria.

Imaging biomarker played an important role in AD detection and its early stage of mild cognitive impairment (MCI) using structural and functional neuroimaging. Recent diagnostic criteria (McKhann et al. 2011) suggest that neuroimaging biomarkers perform better in monitoring AD progression and its early detection than other biomarkers. Different imaging modularities have been used, including computed tomography (CT), magnetic resonance imaging (MRI), functional magnetic resonance imaging (fMRI), positron emission tomography (PET), single photon emission computer tomography (SPECT), diffusion-weighted imaging (DTI), and magnetic resonance spectroscopy (MRS), for biomarkers for detecting AD and predicting its progression. These biomarkers used features such as voxel-wise tissue probability (Kloppel et al. 2008; Liu et al. 2014; Tong et al. 2014), cortical volumes (Wolz et al. 2011; Zhang et al. 2011; Zhou et al. 2011), and thickness (Eskildsen et al. 2013; de Magalhaes Oliveira et al. 2010; Querbes et al. 2009) for classification of AD and MCI.

In this study, we investigate various structural and functional brain imaging techniques to detect AD from HC or MCI individuals. The AD can be detected via its progressive behavior which is illustrated in Fig. 7.1, showing the healthy-control/normal-control (CH/CN), MCI, and AD patient MRI with 2-year follow-up images and their differences. The biomarkers which are automatically extracted and classify AD, HC, and MCI patients are called computer-aided diagnosis (CAD). In this chapter, we discussed the relationship of AD with brain imaging, various types of structure and function neuroimaging techniques, and the observations along with conclusion and future directions.
Fig. 7.1

MRI shows the extracted brain of HC, MCI, and AD patient with 2-year follow-up and their differences (Ledig et al. 2018). (The image is reproduced under the Creative Commons license)

7.2 Relationship of Brain Imaging with AD

AD is a frequent form of neurodegenerative disorder across the world, which has become a public health problem. And imaging plays an important role in the diagnosis of AD and dementia with a special focus on CT, MRI, fMRI, and PET (Johnson et al. 2012). Many of the studies revealed that computer imaging is converted from minor exclusionary role to central position which provides information of temporal and spatial revolution in AD (Selkoe et al. 2012; Krstic and Knuesel 2013). Epidemiological studies indicated that aging and hereditary predisposition are the two key risk factors for AD (Omar et al. 2017). On the other hand, pathological studies of AD have characterized extracellular aggregation of senile plaques (SPs) and development of intracellular neurofibrillary tangles and lesions of cholinergic neurons. Also, molecular pathology such as amyloid deposits can be visualized via imaging in AD (Johnson et al. 2012; Wu et al. 2011). In the central cholinergic system, numerous neurotransmitters and neuronal pathways function together in learning and memory. Thus, the cognitive impairment in AD is related to functional loss in the central cholinergic system (Palle and Neerati 2017). Moreover, aggregation of amyloid-β (Aβ) peptide, which is referred to as amyloid hypothesis, is found to cause synaptic dysfunction and neurodegeneration (van Dyck 2018). These amyloid plaques can’t be seen through structural MRI which cannot detect histopathological trademarks of AD (Johnson et al. 2012). Other pathological processes are also reported in relation to AD such as neuroinflammation, impairment of cerebral circulation, altered synaptic function, and cerebral amyloid angiopathy. Disruption of default network activities during sleeping and resting stage has been revealed via PET imaging (Sperling et al. 2009; Hedden et al. 2009). Hence, these pathological processes that are identified to cause AD are considered important drug targets toward AD drug search. However, still further multi-modality studies are needed to explore the correlationship between MRI, fMRI, fluorodeoxyglucose (FDG)-PET, and PET imaging techniques and AD (Jack et al. 2010; Ghanemi 2015). Considerable number of evidence indicates that dietary control can minimize the risk of developing AD. Animal model studies have also shown a decreasing high-calorie diet to be neuroprotective by reducing Aβ accumulation (Hartman et al. 2006). Generally, flavonoids and in particular quercetin are important compounds for the development of AD therapeutics since it can protect the neurons against oxidative agents and excitotoxicity through regulating cell death mechanisms. FDG-PET is the vital biomarker to detect expression of AD-specific genes, mitochondrial dysfunction, neuropathy, oxidative stress, glial excitation, and synapse loss (Johnson et al. 2012) (Ansari et al. 2009). Oxidative stress reflects an imbalance between reactive oxygen species (ROS) production and biological system’s antioxidant defense mechanisms which act by detoxifying or repairing the reactive intermediates, thereby causing damage. These damages can be seen via PET, ADNI FDG data analysis, fMRI, and MRI (Johnson et al. 2012; Chen et al. 2010; Dong et al. 2014). ROS act as the neurotransmitters and excitatory amino acids in the brain and neuronal tissue. In addition to that, the brain itself represents a substantial source of oxidative stress, as its metabolism serves as a “factory” of ROS which attacks glial cells and neurons, resulting in neuronal damage which may lead in oxidative injury and programmed cell death by apoptosis which is evaluated via computer imaging (Attwell et al. 2010; Hanisch and Kettenmann 2007; Iadecola 2004; Bezzi and Volterra 2001). All these studies revealed that brain imaging/computer imaging is an important tool to explore the histopathological condition of AD. MRI and CT are used to illustrate structural neuroimaging and SPECT, PET, FMRI, and DTI for functional neuroimaging, respectively, which represent the CAD system as shown in Fig. 7.2.
Fig. 7.2

An abstract level of AD classification framework using neuroimaging

7.3 Brain Imaging in AD Diagnosis

7.3.1 Structural Neuroimaging

7.3.1.1 Magnetic Resonance Imaging

Magnetic resonance imaging (MRI) is a medical imaging technique used to measure the hydrogen atoms’ radiofrequency energies. During the static magnetic field, the nuclei with higher energy align against the field, while the lower energy aligns with the field. However, with a specific frequency, the low-energy nuclei absorb energies and align against the field. With the discontinuation of radiation, the MRI scanner detects the emitted energies, and the nuclei return to their lower-energy state. Such changes in energy levels can provide brain structures in the form of detailed images: T1-weighted (T1-w) and T2-weighted (T2-w). T1-w distinguishes gray and white matter, while T2-w characterizes white matter hyperintensities.

MRI is one of the major imaging techniques used to evaluate AD in clinical studies (Byun et al. 2017; Boutet et al. 2014; McEvoy and Brewer 2010) as well as in CAD methods, where computers detect AD automatically using machine learning algorithms (Kloppel et al. 2008; Liu et al. 2014; Tong et al. 2014; Wolz et al. 2011).

Beheshti and Demirel (2016) proposed a feature ranking-based method to classify AD and healthy-control (HC) individuals. Voxel-based morphometry was used to compare the global and local differences, a significant difference was extracted called a volume of interest (VOI), each voxel in VOI is used as a feature, t-test score was used to rank the features, and fisher criterion was used to select the dominant features with support vector machine (SVM) classifier. The work extended (Beheshti et al. 2017) where the same author used the feature ranking based on a genetic algorithm to optimize the feature selection process. In their another work, feature ranking was used with classification error to identify AD patients (Beheshti et al. 2016). Seven different ranking methods were used which include information gain, statistical dependency, t-test score, Pearson correlation coefficient, Gini index, Fisher’s criterion, and mutual information. The classification error was used to select the discriminatory features during the training phase. SVM was used as a classifier with tenfold cross-validation.

Eskildsen et al. (2015) proposed a novel method to predict AD using T1-w MRI. The mutual information fusion was used to identify the robust features which were stable for the period of 3 years. Five features were used to classify AD and MCI subjects which include left and right hippocampi grading, the right anterior part of the parahippocampal gyrus, and the cortical thickness of the left precuneus and left superior temporal sulcus. These features were used in linear discriminant classifier using leave-one-out cross-validation to predict AD and MCI subjects.

Cuingnet et al. (2011) used ten different techniques for CN, AD, and MCI classification. These methods include five voxel-based methods, two methods using the hippocampus, and three methods based on cortical thickness. The results were compared: CN vs AD, CN vs MCI, and MCI non-converted vs MCI converted. This work also evaluated the time complexity of hyper-parameter tuning and feature selection. Westman et al. (2012) used the combination of MRI and cerebrospinal fluid (CSF) measures to detect AD and MCI conversion. In this study, 34 cortical thickness, 34 subcortical volumetric, and 3 CSF measures with a total of 699 (AD = 187, MCI = 287 and CTL = 225) measures were used for classification. The experiments were performed on AD vs HC, MCI vs HC, and MCI vs AD prediction.

Liu et al. (2013) used an unsupervised method, the local linear embedding (LLE) algorithm. The algorithm is used to transform the MRI data to local linear space with few dimensions. These features were used to train a classifier for AD prediction. The work demonstrated that LLE feature performed better in classification than using direct features. Moradi et al. (2015) used the semi-supervised method to detect the conversion of MCI to AD in MCI subjects. The low-density separation method was used to extract features and integrated with cognitive and age measures to improve the performance. The results showed that the combination of these features improved the area under the curve (AUC).

Sørensen et al. (2017) used cortical thickness hippocampal texture, shape, and volumetry as features to detect CN, AD, and MCI patients. The work secured first place in CAD-dementia challenge where it evaluated on two datasets, Alzheimer’s Disease Neuroimaging Initiative (ADNI) and Australian Imaging Biomarkers ad Lifestyle Study of Ageing (AIBL), with tenfold cross-validation. Linear discriminant analysis was used for feature reduction.

7.3.1.2 CT

CT is a diagnostic imaging test which draws the detailed images of internal organs such as bone, vessels, and soft tissues. It is a computer-processed combinations of many X-ray imaging procedure where the rotated beam of X-rays is pointed to the body to make cross-sectional images, also called slices, which can be visualized in three dimensions (3D). The attenuation of the body tissues varies which is why the highest attenuation appears in white such as bones, while other tissues appear in black. That’s why CT has the ability to visualize fluid, gas, and soft tissues. However, it cannot differentiate the gray and white matter (G and WM). To detect AD in CT images, boundary detection is one of the important methods which can increase the detection rate. Al-Jibory and El-Zaart (2013) proposed a method based on Weibull distribution instead of Gaussian distribution to detect boundary detection. The advantages of this method are to address more challenging problem where the distribution has both symmetric and asymmetric shapes.

7.3.2 Functional Neuroimaging

7.3.2.1 SPECT

SPECT is a type of nuclear imaging which is used to view how the blood flows in brain arteries and veins. Radioactive materials are used before SPECT scans, called tracer. These materials emit a single photon which is detected by the machines and translated to two-dimensional cross sections. These cross sections are then converted to 3D brain scans (Schuckit 1992). In comparison with PET, SPECT cannot determine glucose metabolism and also have lower resolution in imaging deep structures. Tabei et al. (2017) used SPECT and white matter hyperintensity (WMH) to predict cognitive decline, which is one of the major problems in AD patients. The study was conducted on 182 patients which concludes that regional cerebral blood flow and WMH are the main parameters that affect the cognitive function in AD patients.

7.3.2.2 PET

PET also uses tracers which emit positron. When a positron encounters with the electron, it gets destroyed and releases a photon which travels in opposite direction. The scanner detects the arrival of two photons simultaneously and determines the travel direction of photons. The scanner then uses this information to contract the PET image just like SPECT (Strauss 1986). However, the PET images have good spatial resolution and are used widely because of lower scanner cost and more availability of PET tracers.

Azmi et al. (2017) used the voxel intensity values of PET images as features to detect AD and HC individuals. The voxel intensities feature consists of mean voxel intensity, slice-based intensity, and global mean voxel intensity. Neural network with nodes 1 and 10 was used for the classification with z-scores. Silveira and Marques (2010) used voxel intensity with a total of 150 features with a weak classifier which combines the outputs to make a complex classifier. Each time the classifier was boosted via re-weighting which improves the performance. The work used AdaBoost algorithm, and for evaluation on selected features, SVM with radial basis function (RBF) kernel was used with tenfold cross-validation. Garali et al. (2015) extracted features from the region of interest and used separation power factor method to select the best 21 regions for further classification which improve the performance and reduce the time complexity. SVM and random forest were used as classifiers. Gray et al. (2012) used multiregional features extracted from PET images. The statistical analysis t-test was conducted to evaluate the proposed method, and two-class SVM was used for classification.

Darsana and Abraham (2016) applied the multistructure registration approach using both PET and MRI images to detect AD and MCI individuals. Asim et al. (2018) applied multi-model and multi-atlas-based method for AD detection using MRI and PET images. Voxel-based measure (VBM), GM, and WM maps were extracted from MRI, while cerebral metabolic rates of glucose were extracted from PET images and used as features. Principal component analysis (PCA) was used to reduce the image features with help of four major steps; normalize image data, calculate covariance matrix from the image data, perform single data decomposition (SVD) and find the projection image data to the new basis with reduced features. SVM with RBF kernel was used for leave-one-out cross-validation. Ota et al. (2015) used Loni probabilistic brain atlas and automated anatomical labeling to extract relative cerebral metabolic rate and gray matter density features. The features were fed to SVM-recursive feature elimination (SVM-RFE) for feature reduction to avoid overfitting and for classification with leave-one-out cross-validation.

7.3.2.3 fMRI

fMRI is used to detect brain activity and changes in blood oxygenation and flow which arise due to the initiation of the activity, also called hemodynamics. The active area of the brain consumes more oxygen which increases blood demand. fMRI produce the activation maps which are represented in different color codes, showing the strength and involvement of specific brain region in a particular mental process. The advantage of this imaging technique is that it does not use any radiation just like CT and PET.

Long et al. (2016) used to detect MCI using SVM-based methods with the multilevel characteristic of MRI. The MCI is the transition phase from CH/HC to AD and an important step for early therapeutic interventions. Hurst exponents, gray matter density, the amplitude of low-frequency fluctuations, and regional homogeneity were used as the main features for SVM with leave-one-out cross-validation. Challis et al. (2015) utilized Gaussian process logistic regression (GP-LR) with SVM using fMRI images to detect CH vs amnestic MCI (MCI-A) and MCI-A vs AD. The number of features was selected via Kendall tau correlation coefficient ranking, and features normalized via PCA and the GP-LR parameters were optimized to increase the accuracy. Leave-one-out cross-validation was used for GP-LR and SVM models.

Khazaee et al. (2017) used local and global graph measure features with SVM and naïve Bayes classifier. Fisher algorithm was used for feature selection based on the discrimination ability, followed by forward-sequential feature selection and tenfold cross-validation for robust classification. Using both the classifiers with optimal features, Bayes classifier performed better using all patient data and achieved higher AUC in HC vs AD classification. Hojjati et al. (2017) extracted graph measures of segregation and integration using graph theory. The study focused on MCI patients where they classify MCI converter (MCI-C) from MCI non-converter (MCI-NC). Global and local graph measures were calculated, which ended with 913 features. Feature selection method was used to sort the features, and sequential feature collection was used to find the optimal features for classification. SVM with leave one out cross-validation was used along the statistical analysis.

Schouten et al. (2016) combined various modalities of MRI and fMRI to classify AD and HC individuals. Various features were extracted including anatomical, diffusion, and functional connectivity features with the elastic net classifier using tenfold cross-validation. In another study (de Vos et al. 2018), eight measures were extracted including functional connectivity (FC) dynamics, FC metric, FC states, graph properties, FC for each voxel, FC for each hippocampus and low-frequency fluctuation, eigenvector centrality, and the combination of all these features. Agosta et al. (2012) explored the salience, default mode, and executive networks to detect the resting-state abnormality in MCI-A and HC. The experiments were conducted on very few images which is one of the limitations of this study. Li et al. (2018) used a transfer learning approach to improve the detection rate on a smaller dataset. ADNI dataset was used for training, and then the trained weights were transferred via adaptation method for Tongji dataset classification which consists of only 12 AD and 14 CN individual’s data. The results improved by 30% than that using the smaller dataset.

7.3.2.4 DTI

DTI is the extension of diffusion-weighted imaging which provides data about WM region orientation. The measurements are based on Brownian motion and water molecules. DTI provides quantitative information to track the magnitude, orientation, and anisotropy of brain WM regions with the help of computer algorithms.

Ahmed et al. (2017) developed a CAD system by extracting features from T1-w MRI and mean diffusivity from DTI images. The AD disease-related signature was generated which was used in multiple kernel learning approach for classification with tenfold cross-validation. Kantarci et al. (2017) used DTI images in clinical perspective to analyze the white matter integrity and pathologic staging in AD. The voxel-based and atlas-based analyses were used for the evaluation of neurofibrillary tangle stages. Schouten et al. (2017) used voxel tensor measures, graph measures, and independent component analysis (ICA)-clustered measures. All features were used for classification independently and also as a fusion vector via elastic net classification with tenfold cross-validation. The results demonstrate that fractional anisotropy with ICA performed better than other measures.

7.3.2.5 MRS

MRS is considered as a noninvasive diagnostic test to estimate the chemical changes in the brain. MRS compares the normal and abnormal tissues, while MRI is used to detect the anatomical location of abnormal tissues. Most commonly used MRS is proton spectroscopy which analyzes the proton molecules. MRS is used to quantify metabolites such as alanine, N-acetyl-aspartate (NAA), creatine (CR), amino acids, and myoinositol.

Proton MRS was used by Wang et al. (2015) to identify the cerebral metabolite changes in AD patients, which shows that NAA was significantly reduced in bilateral hippocampus and posterior cingulate. NAA/CR ratio also decreased in posterior cingulate. Other studies also discovered the reduction of NAA/CR in AD patients (Schuff et al. 1997; Adalsteinsson et al. 2000).

7.4 Observations

In this study, we observed that most of the biomarkers are extracted using MRI, fMRI, and PET, while the rest of the images such as DTI are used by very few studies as shown in Table 7.1. It is well understood that MRI, fMRI, and PET images separately and in combination provide more details to extract biomarkers for AD with better classification rate as shown in Table 7.1. However, it’s unfair to compare the ACC, SEN, and specificity (SPE) of MRI, fMRI, PET, and DTI listed papers because they are not using the same classification pairs or the same datasets, i.e., Beheshti and Demirel (2016) classify CN and AD using MRIs of ADNI dataset, while Garali et al. (2015) classify CN and AD using PET but with a private dataset. Similarly, Silveira and Marques (2010) used PET and ADNI dataset but did the classification of AD and MCI subjects. Table 7.1 also shows that only a few studies used unsupervised or semi-supervised methods (Liu et al. 2013; Moradi et al. 2015), while the rest of the methods used a supervised method with naïve Bayes (Khazaee et al. 2017), neural network (Azmi et al. 2017), and SVM (Long et al. 2016; Hojjati et al. 2017) classifiers.
Table 7.1

Summary of AD detection methods using brain imaging

References

Images

Methods

Participants

Results (%)

Datasets

Healthy-control (HC)/normal-control (CN)

Alzheimer’s disease (AD)

Mild cognitive impairment (MCI)

Beheshti and Demirel (2016)

MRI

Feature ranking-based approach

68

68

ACC = 96.32

SEN = 94.11

SPE = 98.52

AUC = 0.9993

ADNI (Wyman et al. 2013)

Beheshti et al. (2017)

MRI

Feature ranking + genetic algorithm

162

160

MCI-S = 65

MCI-P = 71

ACC = 93.01

SEN = 89.13

SPE = 96.80

AUC = 0.9351

ADNI

Eskildsen et al. (2015)

MRI (T1-w)

Measurements of structural pathologic patterns

231

198

405

ACC = 71.9

SEN = 69.6

SPE = 73.6

AUC = 0.763

ADNI

Beheshti et al. (2016)

MRI

Feature ranking + classification error

130

130

ACC = 92.38

SEN = 91.07

SPE = 93.89

AUC = 0.96

ADNI

Cuingnet et al. (2011)

MRI

Comparative analysis of 10 methods

81

69

MCI-NC = 67

MCI-C = 37

For CN vs AD:

SEN = 81

SPE = 95

ADNI

Westman et al. (2012)

MRI

MRI measures with combination of CSF measures

111

96

162

ACC = 91.8

SEN = 88.5

SPE = 94.6

AUC = 0.958

ADNI

Liu et al. (2013)

MRI

LLE

137

96

MCI-S = 93

MCI-C = 97

ACC = 68.0

SEN = 80.0

SPE = 56.0

P = 0.007

ADNI

Moradi et al. (2015)

MRI

Low-density separation

231

200

MCI-P = 164

MCI-S = 100

AUC = 0.902

ADNI

Sørensen et al. (2017)

MRI (T1-w)

Cortical thickness, hippocampal shape, texture, and volumetry

ADNI: 169

AIBL: 88

CAD-dementia:12

ADNI: 101

AIBL: 28

CAD-dementia:9

ADNI: 234

AIBL: 29

CAD-dementia:9

ACC = 63.0

AUC = 0.832

ADNI, AIBL (Ellis et al. 2009), CAD-dementia (Bron et al. 2015)

Azmi et al. (2017)

PET

Voxel intensities + neural network

219

126

Nural-net 10 Nodes:

SEN = 83.0

SPE = 94.0

Nural-Net 1 node:

ACC = 90.0

SEN = 81.0

SPE = 95.0

ADNI

Gray et al. (2012)

PET

Multiregional analysis

54

50

MCI-S = 64

MCI-P = 53

ACC = 88.4

SEN = 83.2

SPE = 93.6

AUC = 0.94

ADNI

Silveira and Marques (2010)

PET

Voxel intensities + AdaBoost

81

74

113

AD detection:

ACC = 90.97

MCI detection:

ACC = 79.63

ADNI

Garali et al. (2015)

PET

Region-based features

61

81

ACC = 95.07

“La-Timone” University Hospital, in the Nuclear Medicine Department (Marseille, France)

Darsana and Abraham (2016)

PET, MRI

Multi-structure registration approach

133

166

151

ACC = 96.4

ADNI

Asim et al. (2018)

PET, MRI

Multimodal and multi-atlas-based approach

100

100

100

CN vs AD:

ACC = 94.00

SEN = 95.0

SPE = 93.0

ADNI

Ota et al. (2015)

PET, MRI

Relative cerebral metabolic rate + gray matter density + SVM-RFE

40

MCI-A = 40

AUC = 0.75

SEAD-J (Ota et al. 2014)

Zhu et al. 2015 (Dong et al. 2014)

PET, MRI

Sparse multitasking learning framework

52

51

MCI-C = 43

MCI-NC = 56

AD vs NC

ACC = 95.7

SEN = 96.6

SPE = 98.2

AUC = 0.981

ADNI

Long et al. (2016)

fMRI

SVM-based method with multilevel characteristic of MRI

33

29

ACC = 96.77

SEN = 93.10

SPE = 100

Private dataset

Challis et al. (2015)

re-fMRI

Gaussian process logistic regression + SVM

39

27

MCI-A = 50

CN vs MCI-A

ACC = 75%

SEN = 80

SPE = 70

AUC = 0.7

MCI-A vs AD

ACC = 97%

SEN = 90.0

SPE = 90.0

AUC = 0.89

Private Dataset

Hojjati et al. (2017)

re-fMRI

Graph theory + SVM

MCI-C = 18

MCI-NC = 62

MCI-C vs MCI-NC

ROC = 91.4

SEN = 83.24

SPE = 90.1

AUC = 0.95

ADNI

Khazaee et al. (2017)

re-fMRI

Directed graph measures + naïve Bayes classifier

45

34

89

Overall ACC = 93.3

HC vs MCI

AUC = 0.92

MCI vs AD

AUC = 0.89

HC vs AD

AUC = 0.94

ADNI

Agosta et al. (2012)

re-fMRI

Default mode + fontal network + salience networks

13

13

MCI-A = 12

Default mode network I:

F = 19.52

P < 0.001

Executive Network:

F = 11.19

P < 0.001

Outpatient Memory Clinic of the IRCCS Centro San Giovanni di Dio, Brescia, Italy.

de Vos et al. (2018)

re-fMRI

Functional connectivity + independent component analysis

173

77

ACC = 79.0

SEN = 86.0

SPE = 71.0

AUC = 0.85

PRODEM (Seiler et al. 2012)

Schouten et al. (2016)

MRI, re-fMRI

Multi-modalities + elastic net classifier

173

77

ACC = 93.0

SEN = 81.6

SPE = 95.6

AUC = 0.971

PRODEM

Li et al. (2018)

fMRI

Transfer learning

ADNI:175

Tongii:14

ADNI:117

Tongii:12

With adaptation:

ACC = 84.6

SEN = 92.0

SPE = 79.0

AUC = 0.80

ADNI, Tongji Hospital at Wuhan

Ahmed et al. (2017)

DTI, MRI

Multimodal approach + multiple kernel learning

52

45

58

AD vs NC

ACC = 90.2

SEN = 82.92

SPE = 97.2

ADNI

Schouten et al. (2017)

DTI

Voxel-wise measures + elastic net classification

173

77

ACC = 85.1

SEN = 86.8

SPE = 84.4

AUC = 0.92 ± 0.018

PRODEM

Some of the notations used in this table are accuracy (ACC), sensitivity (SEN), specificity (SPE), area under the curve (AUC), F-score (F), and p-value (P)

Deep learning (DL) is another supervised method which recently got attention due to its state-of-the-art performance in image processing and computer vision. DL also cultivated the area of medical imaging (Klang 2018; Lakhani et al. 2018); the auto-encoder, deep neural networks, and the most widely used 2D and 3D convolutional neural networks (CNN) are focused on detection, segmentation, registrations, content-based image retrieval, image reconstruction, and generation (Litjens et al. 2017). However, in the area of AD detection, DL work is limited and it may be due to the small amount of data. DL algorithms such as CNN need a large amount of data, resources, and time to train the model and tune the hyperparameters. To all such issues, one solution is transfer learning, train the model on large dataset, and transfer the trained weights for small dataset image classification as adopted by Li et al. (2018) for AD small dataset classification. Most of the studies conducted the experiments on ADNI (Wyman et al. 2013) dataset which is the largest AD dataset publicly available. According to ADNI, a total of 800 participants including 200 CH/HC subjects, 400 MCI individuals, and 200 AD subjects are recruited from 50 different places from the United States and Canada.

The evaluation metrics for biomarker used in AD are accuracy, SPE, and sensitivity (SEN) which shows the true-positive (TP) rate and true-negative (TN) rate, respectively, while AUC shows the classifier stability, and receiver operating characteristic (ROC) shows the classifier diagnostic capability. These metrics can be extracted from a confusion matrix which consists of TP, TN, false-positive (FP), and false-negative (FN). The mathematical definitions are as follows:

$$ \mathrm{Accuracy}=\frac{\left|\mathrm{TN}\left|+\right|\mathrm{TP}\right|}{\left|\mathrm{TN}\left|+\left|\mathrm{TP}\right|+\left|\mathrm{FP}\right|+\right|\mathrm{FN}\right|} $$
$$ \mathrm{Sensitivity}=\frac{\left|\mathrm{TP}\right|}{\left|\mathrm{TP}\left|+\right|\mathrm{FN}\right|} $$
$$ \mathrm{Specificity}=\frac{\left|\mathrm{TN}\right|}{\left|\mathrm{TN}\left|+\right|\mathrm{FP}\right|} $$

7.5 Conclusion and Future Perspective

AD, one of the leading neurodegenerative disorders, is characterized by brain damage and cognitive impairment in elderly people. Due to the severity of the disease, prevention and/or regression of AD is a challenge to overcome. To track its progression, medical imaging played a vital role from decades, but still, there is room for improvement. The brain images are used to extract the biomarkers which help the doctors in the identification of AD. In this study, we presented various brain image modularities where the research focused more on MRI due to its clarity, availability, and correctly exposing the biomarkers than other imaging techniques (CT, SPECT, PET, FMRI, and DTI).

In future effort, researchers should consider more datasets with a large amount of imaging data for experimentation to improve the AD detection. Researchers should not only focus on imaging but also take clinical symptoms into consideration and used as features which can help in better identification of AD. DL is a new area which has been enriching in medical imaging. However, DL and transfer learning still have not been exploited to the extent in CAD systems for AD.

Notes

Acknowledgments

The authors wish to acknowledge King Fahd University of Petroleum and Minerals (KFUPM) for utilizing the various facilities in carrying out this research. Other authors of the manuscript thank and acknowledge their respective universities and institutes as well.

Conflict of Interest

There is no conflict of interest.

References

  1. 2018 Alzheimer’s disease facts and figures (2018) Alzheimer’s Dement 14(3):367–429Google Scholar
  2. Adalsteinsson E, Sullivan EV, Kleinhans N, Spielman DM, Pfefferbaum A (2000) Longitudinal decline of the neuronal marker N-acetyl aspartate in Alzheimer’s disease. Lancet (London, England) 355(9216):1696–1697CrossRefGoogle Scholar
  3. Agosta F, Pievani M, Geroldi C, Copetti M, Frisoni GB, Filippi M (2012) Resting state fMRI in Alzheimer’s disease: beyond the default mode network. Neurobiol Aging 33(8):1564–1578PubMedCrossRefGoogle Scholar
  4. Al-Jibory WK, El-Zaart A (2013) Edge detection for diagnosis early Alzheimer’s disease by using Weibull distribution. In: 2013 25th International Conference on Microelectronics (ICM), pp 1–5Google Scholar
  5. Alzheimer’s Association (2014) 2014 Alzheimer’s disease facts and figures. Alzheimers Dement 10(2):e47–e92CrossRefGoogle Scholar
  6. Ansari MA, Abdul HM, Joshi G, Opii WO, Butterfield DA (2009) Protective effect of quercetin in primary neurons against Aβ(1–42): relevance to Alzheimer’s disease. J Nutr Biochem 20(4):269–275PubMedCrossRefGoogle Scholar
  7. Asim Y, Raza B, Malik AK, Rathore S, Hussain L, Iftikhar MA (2018) A multi-modal, multi-atlas-based approach for Alzheimer detection via machine learning. Int J Imaging Syst Technol 28(2):113–123CrossRefGoogle Scholar
  8. Attwell D, Buchan AM, Charpak S, Lauritzen M, MacVicar BA, Newman EA (2010) Glial and neuronal control of brain blood flow. Nature 468(7321):232–243PubMedPubMedCentralCrossRefGoogle Scholar
  9. Azmi MH, Saripan MI, Nordin AJ, Ahmad Saad FF, Abdul Aziz SA, Wan Adnan WA (2017) 18 F-FDG PET brain images as features for Alzheimer classification. Radiat Phys Chem 137:135–143CrossRefGoogle Scholar
  10. Beheshti I, Demirel H (2016) Feature-ranking-based Alzheimer’s disease classification from structural MRI. Magn Reson Imaging 34(3):252–263PubMedCrossRefGoogle Scholar
  11. Beheshti I, Demirel H, Farokhian F, Yang C, Matsuda H (2016) Structural MRI-based detection of Alzheimer’s disease using feature ranking and classification error. Comput Methods Prog Biomed 137:177–193CrossRefGoogle Scholar
  12. Beheshti I, Demirel H, Matsuda H (2017) Classification of Alzheimer’s disease and prediction of mild cognitive impairment-to-Alzheimer’s conversion from structural magnetic resource imaging using feature ranking and a genetic algorithm. Comput Biol Med 83:109–119PubMedCrossRefGoogle Scholar
  13. Ben Ahmed O, Benois-Pineau J, Allard M, Catheline G, Ben Amar C (2017) Recognition of Alzheimer’s disease and mild cognitive impairment with multimodal image-derived biomarkers and Multiple Kernel learning. Neurocomputing 220:98–110CrossRefGoogle Scholar
  14. Bezzi P, Volterra A (2001) A neuron-glia signalling network in the active brain. Curr Opin Neurobiol 11(3):387–394PubMedCrossRefGoogle Scholar
  15. Boutet C et al (2014) Detection of volume loss in hippocampal layers in Alzheimer’s disease using 7 T MRI: a feasibility study. NeuroImage Clin 5:341–348PubMedPubMedCentralCrossRefGoogle Scholar
  16. Bron EE et al (2015) Standardized evaluation of algorithms for computer-aided diagnosis of dementia based on structural MRI: The CAD dementia challenge. NeuroImage 111:562–579PubMedPubMedCentralCrossRefGoogle Scholar
  17. Byun MS et al (2017) Korean brain aging study for the early diagnosis and prediction of Alzheimer’s disease: methodology and baseline sample characteristics. Psychiatry Investig 14(6):851PubMedPubMedCentralCrossRefGoogle Scholar
  18. Challis E, Hurley P, Serra L, Bozzali M, Oliver S, Cercignani M (2015) Gaussian process classification of Alzheimer’s disease and mild cognitive impairment from resting-state fMRI. NeuroImage 112:232–243PubMedCrossRefGoogle Scholar
  19. Chen K et al (2010) Twelve-month metabolic declines in probable Alzheimer’s disease and amnestic mild cognitive impairment assessed using an empirically pre-defined statistical region-of-interest: findings from the Alzheimer’s disease neuroimaging initiative. NeuroImage 51(2):654–664PubMedPubMedCentralCrossRefGoogle Scholar
  20. Cuingnet R et al (2011) Automatic classification of patients with Alzheimer’s disease from structural MRI: a comparison of ten methods using the ADNI database. NeuroImage 56(2):766–781PubMedCrossRefGoogle Scholar
  21. Darsana S, Abraham L (2016) Multistructure brain registration using multimodal neuroimaging for the detection of Alzheimer’s disease. In: 2016 International Conference on Communication Systems and Networks (ComNet), pp 221–226Google Scholar
  22. de Magalhaes Oliveira PP, Nitrini R, Busatto G, Buchpiguel C, Sato JR, Amaro E (2010) Use of SVM methods with surface-based cortical and volumetric subcortical measurements to detect Alzheimer’s disease. J Alzheimers Dis 19(4):1263–1272CrossRefGoogle Scholar
  23. de Vos F et al (2018) A comprehensive analysis of resting state fMRI measures to classify individual patients with Alzheimer’s disease. NeuroImage 167:62–72PubMedCrossRefGoogle Scholar
  24. Dong Y et al (2014) Protective effect of quercetin against oxidative stress and brain edema in an experimental rat model of subarachnoid hemorrhage. Int J Med Sci 11(3):282–290PubMedPubMedCentralCrossRefGoogle Scholar
  25. Ellis KA et al (2009) The Australian Imaging, Biomarkers and Lifestyle (AIBL) study of aging: methodology and baseline characteristics of 1112 individuals recruited for a longitudinal study of Alzheimer’s disease. Int Psychogeriatr 21(4):672PubMedCrossRefGoogle Scholar
  26. Eskildsen SF, Coupé P, García-Lorenzo D, Fonov V, Pruessner JC, Collins DL (2013) Prediction of Alzheimer’s disease in subjects with mild cognitive impairment from the ADNI cohort using patterns of cortical thinning. NeuroImage 65:511–521PubMedCrossRefGoogle Scholar
  27. Eskildsen SF, Coupé P, Fonov VS, Pruessner JC, Collins DL (2015) Structural imaging biomarkers of Alzheimer’s disease: predicting disease progression. Neurobiol Aging 36:S23–S31PubMedCrossRefGoogle Scholar
  28. Garali I, Adel M, Bourennane S, Guedj E (2015) Region-based brain selection and classification on pet images for Alzheimer’s disease computer aided diagnosis. In: 2015 IEEE International Conference on Image Processing (ICIP), pp 1473–1477Google Scholar
  29. Ghanemi A (2015) Alzheimer’s disease therapies: Selected advances and future perspectives. Alexandria J Med 51(1):1–3CrossRefGoogle Scholar
  30. Gray KR, Wolz R, Heckemann RA, Aljabar P, Hammers A, Rueckert D (2012) Multi-region analysis of longitudinal FDG-PET for the classification of Alzheimer’s disease. NeuroImage 60(1):221–229PubMedPubMedCentralCrossRefGoogle Scholar
  31. Hanisch U-K, Kettenmann H (2007) Microglia: active sensor and versatile effector cells in the normal and pathologic brain. Nat Neurosci 10(11):1387–1394PubMedCrossRefGoogle Scholar
  32. Hartman RE et al (2006) Pomegranate juice decreases amyloid load and improves behavior in a mouse model of Alzheimer’s disease. Neurobiol Dis 24(3):506–515PubMedCrossRefGoogle Scholar
  33. Hedden T et al (2009) Disruption of functional connectivity in clinically normal older adults harboring amyloid burden. J Neurosci 29(40):12686–12694PubMedPubMedCentralCrossRefGoogle Scholar
  34. Hojjati SH, Ebrahimzadeh A, Khazaee A, Babajani-Feremi A (2017) Predicting conversion from MCI to AD using resting-state fMRI, graph theoretical approach and SVM. J Neurosci Methods 282:69–80PubMedCrossRefGoogle Scholar
  35. Iadecola C (2004) Neurovascular regulation in the normal brain and in Alzheimer’s disease. Nat Rev Neurosci 5(5):347–360PubMedCrossRefGoogle Scholar
  36. Jack CR et al (2010) Hypothetical model of dynamic biomarkers of the Alzheimer’s pathological cascade. Lancet Neurol 9(1):119–128PubMedPubMedCentralCrossRefGoogle Scholar
  37. Johnson KA, Fox NC, Sperling RA, Klunk WE (2012) Brain imaging in Alzheimer disease. Cold Spring Harb Perspect Med 2(4):a006213–a006213PubMedPubMedCentralCrossRefGoogle Scholar
  38. Kantarci K et al (2017) White-matter integrity on DTI and the pathologic staging of Alzheimer’s disease. Neurobiol Aging 56:172–179PubMedPubMedCentralCrossRefGoogle Scholar
  39. Khazaee A, Ebrahimzadeh A, Babajani-Feremi A, Alzheimer’s Disease Neuroimaging Initiative (2017) Classification of patients with MCI and AD from healthy controls using directed graph measures of resting-state fMRI. Behav Brain Res 322(Pt B):339–350PubMedCrossRefPubMedCentralGoogle Scholar
  40. Klang E (2018) Deep learning and medical imaging. J Thorac Dis 10(3):1325–1328PubMedPubMedCentralCrossRefGoogle Scholar
  41. Kloppel S et al (2008) Automatic classification of MR scans in Alzheimer’s disease. Brain 131(3):681–689PubMedPubMedCentralCrossRefGoogle Scholar
  42. Krstic D, Knuesel I (2013) Deciphering the mechanism underlying late-onset Alzheimer disease. Nat Rev Neurol 9(1):25–34PubMedCrossRefGoogle Scholar
  43. Lakhani P, Gray DL, Pett CR, Nagy P, Shih G (2018) Hello world deep learning in medical imaging. J Digit Imaging 31(3):283–289PubMedPubMedCentralCrossRefGoogle Scholar
  44. Ledig C, Schuh A, Guerrero R, Heckemann RA, Rueckert D (2018) Structural brain imaging in Alzheimer’s disease and mild cognitive impairment: biomarker analysis and shared morphometry database. Sci Rep 8(1):11258PubMedPubMedCentralCrossRefGoogle Scholar
  45. Li W, Zhao Y, Chen X, Xiao Y, Qin Y (2018) Detecting Alzheimer’s disease on small dataset: a knowledge transfer perspective. IEEE J Biomed Heal Informatics:1–1Google Scholar
  46. Litjens G et al (2017) A survey on deep learning in medical image analysis. Med Image Anal 42:60–88PubMedCrossRefGoogle Scholar
  47. Liu X, Tosun D, Weiner MW, Schuff N (2013) Locally linear embedding (LLE) for MRI based Alzheimer’s disease classification. NeuroImage 83:148–157PubMedCrossRefGoogle Scholar
  48. Liu M, Zhang D, Shen D (2014) Hierarchical fusion of features and classifier decisions for Alzheimer’s disease diagnosis. Hum Brain Mapp 35(4):1305–1319PubMedCrossRefGoogle Scholar
  49. Long Z et al (2016) A support vector machine-based method to identify mild cognitive impairment with multi-level characteristics of magnetic resonance imaging. Neuroscience 331:169–176PubMedCrossRefGoogle Scholar
  50. McEvoy LK, Brewer JB (2010) Quantitative structural MRI for early detection of Alzheimer’s disease. Expert Rev Neurother 10(11):1675–1688PubMedPubMedCentralCrossRefGoogle Scholar
  51. McKhann GM et al (2011) The diagnosis of dementia due to Alzheimer’s disease: recommendations from the National Institute on Aging-Alzheimer’s Association workgroups on diagnostic guidelines for Alzheimer’s disease. Alzheimers Dement 7(3):263–269PubMedPubMedCentralCrossRefGoogle Scholar
  52. Moradi E, Pepe A, Gaser C, Huttunen H, Tohka J (2015) Machine learning framework for early MRI-based Alzheimer’s conversion prediction in MCI subjects. NeuroImage 104:398–412PubMedCrossRefGoogle Scholar
  53. Omar SH, Scott CJ, Hamlin AS, Obied HK (2017) The protective role of plant biophenols in mechanisms of Alzheimer’s disease. J Nutr Biochem 47:1–20PubMedCrossRefGoogle Scholar
  54. Ota K, Oishi N, Ito K, Fukuyama H (2014) A comparison of three brain atlases for MCI prediction. J Neurosci Methods 221:139–150PubMedCrossRefGoogle Scholar
  55. Ota K, Oishi N, Ito K, Fukuyama H (2015) Effects of imaging modalities, brain atlases and feature selection on prediction of Alzheimer’s disease. J Neurosci Methods 256:168–183PubMedCrossRefGoogle Scholar
  56. Palle S, Neerati P (2017) Quercetin nanoparticles attenuates scopolamine induced spatial memory deficits and pathological damages in rats. Bull Fac Pharmacy, Cairo Univ 55(1):101–106CrossRefGoogle Scholar
  57. Querbes O et al (2009) Early diagnosis of Alzheimer’s disease using cortical thickness: impact of cognitive reserve. Brain 132(8):2036–2047PubMedPubMedCentralCrossRefGoogle Scholar
  58. Schouten TM et al (2016) Combining anatomical, diffusion, and resting state functional magnetic resonance imaging for individual classification of mild and moderate Alzheimer’s disease. NeuroImage Clin 11:46–51PubMedPubMedCentralCrossRefGoogle Scholar
  59. Schouten TM et al (2017) Individual classification of Alzheimer’s disease with diffusion magnetic resonance imaging. NeuroImage 152:476–481PubMedCrossRefGoogle Scholar
  60. Schuckit MA (1992) An introduction and overview to clinical applications of neuroSPECT in psychiatry. J Clin Psychiatry 53(Suppl):3–6PubMedPubMedCentralGoogle Scholar
  61. Schuff N et al (1997) Changes of hippocampal N-acetyl aspartate and volume in Alzheimer’s disease. A proton MR spectroscopic imaging and MRI study. Neurology 49(6):1513–1521PubMedCrossRefGoogle Scholar
  62. Seiler S et al (2012) Driving Cessation and dementia: results of the prospective registry on dementia in Austria (PRODEM). PLoS One 7(12):e52710PubMedPubMedCentralCrossRefGoogle Scholar
  63. Selkoe D, Mandelkow E, Holtzman D (2012) Deciphering Alzheimer disease. Cold Spring Harb Perspect Med 2(1):a011460–a011460PubMedPubMedCentralCrossRefGoogle Scholar
  64. Silveira M, Marques J (2010) Boosting Alzheimer disease diagnosis using PET images. In: 2010 20th International Conference on Pattern Recognition, pp 2556–2559Google Scholar
  65. Sørensen L et al (2017) Differential diagnosis of mild cognitive impairment and Alzheimer’s disease using structural MRI cortical thickness, hippocampal shape, hippocampal texture, and volumetry. NeuroImage Clin 13:470–482PubMedCrossRefGoogle Scholar
  66. Sperling RA et al (2009) Amyloid deposition is associated with impaired default network function in older persons without dementia. Neuron 63(2):178–188PubMedPubMedCentralCrossRefGoogle Scholar
  67. Strauss HW (1986) Positron emission tomography and autoradiography: principles and applications for the brain and heart. Phelps ME, Mazziotta JC, Schelbert HR (eds) Raven Press, New York, 704 Pages, Illustrated, $89.50 ISBN: 0-88167-118-5. Clin Cardiol 9(5):233–233Google Scholar
  68. Tabei K, Kida H, Hosoya T, Satoh M, Tomimoto H (2017) Prediction of cognitive decline from white matter hyperintensity and single-photon emission computed tomography in Alzheimer’s disease. Front Neurol 8:408PubMedPubMedCentralCrossRefGoogle Scholar
  69. Tong T, Wolz R, Gao Q, Guerrero R, Hajnal JV, Rueckert D (2014) Multiple instance learning for classification of dementia in brain MRI. Med Image Anal 18(5):808–818PubMedCrossRefGoogle Scholar
  70. van Dyck CH (2018) Anti-amyloid-β monoclonal antibodies for Alzheimer’s disease: pitfalls and promise. Biol Psychiatry 83(4):311–319PubMedCrossRefGoogle Scholar
  71. Wang H et al (2015) Magnetic resonance spectroscopy in Alzheimer’s disease: systematic review and meta-analysis. J Alzheimers Dis 46(4):1049–1070PubMedCrossRefGoogle Scholar
  72. Westman E, Muehlboeck J-S, Simmons A (2012) Combining MRI and CSF measures for classification of Alzheimer’s disease and prediction of mild cognitive impairment conversion. NeuroImage 62(1):229–238PubMedCrossRefGoogle Scholar
  73. Wolz R et al (2011) Multi-method analysis of MRI images in early diagnostics of Alzheimer’s disease. PLoS One 6(10):e25446PubMedPubMedCentralCrossRefGoogle Scholar
  74. Wu T-Y, Chen C-P, Jinn T-R (2011) Traditional Chinese medicines and Alzheimer’s disease. Taiwan J Obstet Gynecol 50(2):131–135PubMedCrossRefGoogle Scholar
  75. Wyman BT et al (2013) Standardization of analysis sets for reporting results from ADNI MRI data. Alzheimers Dement 9(3):332–337PubMedCrossRefGoogle Scholar
  76. Zhang D, Wang Y, Zhou L, Yuan H, Shen D (2011) Multimodal classification of Alzheimer’s disease and mild cognitive impairment. NeuroImage 55(3):856–867PubMedPubMedCentralCrossRefGoogle Scholar
  77. Zhou L, Wang Y, Li Y, Yap P-T, Shen D (2011) Hierarchical anatomical brain networks for MCI prediction: revisiting volumetric measures. PLoS One 6(7):e21935PubMedPubMedCentralCrossRefGoogle Scholar

Copyright information

© Springer Nature Singapore Pte Ltd. 2019

Authors and Affiliations

  1. 1.Department of Information and Computer ScienceKing Fahd University of Petroleum and MineralsDhahranKingdom of Saudi Arabia
  2. 2.Department of Pharmacology and Toxicology, Faculty of Bio-SciencesCholistan University of Veterinary and Animal Sciences (CUVAS)BahawalpurPakistan
  3. 3.Neurosciences CenterKing Fahd Specialist Hospital DammamDammamKingdom of Saudi Arabia

Personalised recommendations