Introduction

Alzheimer’s disease (AD) is a progressive neurodegenerative disease without any certain treatment until now and leads to death eventually. In addition, AD mainly affects older people over the age of 65 years with an exponentially increasing rate, nearly doubling every five years [ 1 ]. However, Alzheimer’s has no definitive cure [ 2 , 3 ], and the detection of the disease in the early stage can enormously assist in slowing down the progress, leading to effective treatment.

Some tests are used to diagnose Alzheimer’s, such as mini-mental exams [ 4 ], distinguishing the cognitive symptoms of the disease, and brain imaging techniques, such as Magnetic Resonance Imaging (MRI) [ 5 - 8 ]. The neuropathological alteration due to AD can appear much earlier before the onset of clinical symptoms [ 9 ]. Therefore, the early detection of AD using neuroimaging techniques is considered a promising area of research, especially with the advances in machine-learning and image-segmentation techniques [ 10 - 16 ].

MRI scans have been investigated to obtain many Alzheimer’s biomarkers and study the most atrophic regions using volume measurements [ 6 , 17 ], shape [ 18 ], texture [ 17 , 19 , 20 ], cortical measurements [ 21 , 22 ], and sulcal measurements [ 23 ]. These measurements were applied to many brain regions, such as the hippocampus [ 24 ], which is one of the earliest brain regions in the neurodegeneration [ 25 ], amygdala [ 26 , 27 ], whole brain [ 28 ], entorhinal cortex [ 29 ], brainstem [ 30 ], and ventricles [ 31 ]. Recent advances in machine-learning techniques, such as Support Vector Machine (SVM) [ 32 , 33 ], Naïve Bayes, Logistic Regression, and K-Nearest Neighbors (KNN ) [ 34 ] have been implemented. The use of automated methods rather than relying solely on physician experiments has led to the reliance on ensemble models to improve disease detection and increase accuracy. However, a major challenge is in selecting the best biomarkers that characterize AD to differentiate between AD and Normal Controls (NC).

Several feature selection methods have been used in recent studies; for example, Particle Swarm Optimization (PSO) algorithm [ 35 ], genetic algorithm, t-test [ 36 , 37 ], and Principal Component Analysis (PCA) [ 38 - 41 ].

The current study aimed to demonstrate the least and most beneficial number of features among a large pool of different AD biomarkers to classify AD cases and perform the best classifiers using these features.

Material and Methods

Database

In this analytical study, data were acquired from the Alzheimer’s disease Neuroimaging Initiative (ADNI) database (http://adni.loni.usc.edu), propelled as a public-private corporation by six nonprofit organizations in 2003 as follows: the National Institute on Aging (NIA), the National Institute of Biomedical Imaging and Bioengineering (NIBIB), the Food and Drug Administration (FDA), and private pharmaceutical companies. ADNI’s main objective was to check whether some specific biomarkers, clinical and neuropsychological assessment, positron emission tomography (PET), and serial MRI can be combined to evaluate the Mild Cognitive Impairment (MCI) evolution and early Alzheimer’s.

154 T1-weighted images were obtained from ADNI, 37 female cases and 41 male cases in the AD stage, and 40 females and 36 males in the normal control (NC) stage. The age ranged from 50 to 85 years. The magnetic field strength was 3T, slice thickness was 1.2 mm, acquisition matrix was 240×256 pixels with pixel spacing X=1.0 mm; pixel spacing Y=1.0 mm, the number of slices=176, and demographic characteristics of the individuals (shown in Table 1).

Image Preprocessing

Before executing the analysis, the quality of the data must be improved due to missing values and inaccurate information, leading to distorted results. The data was preprocessed using CAT12 after obtaining it from ADNI. The preprocessing workflow involved bias field inhomogeneities correction, affine registration, skull stripping, and normalization to Montreal Neurological Institute (MNI). Hammers atlas [ 42 ] is then used as a binary mask to select the brain Regions of Interest (ROIs), as shown in Figure 1. Finally, 71 raw volumetric measurements, 68 cortical thickness (CT), 68 gyrification indexes (GI), and 68 sulcal depth (SD) measurements are extracted. The four measurements, together with their differences among AD and NC, are shown in Figure 2. Volume measurements, involving the hippocampus, amygdala, temporal pole, fusiform, insula, putamen, thalamus, lateral temporal ventricle, and cuneus were normalized by the intracranial volume. Relative volumes provide more precise volumes to reduce the influence of factors, such as the head and brain size. Entorhinal, temporal pole, fusiform, parahippocampus, and insula are examples of surface-based characteristics (CT, GI, and SD), resulting in the excellent features to indicate the existence of the disease.

Figure 1. Workflow of medical image preprocessing

Figure 2. Brain mapping of cortical thickness, gyrification index, and sulcus depth maps estimated using CAT12 toolbox. Each column denotes a subject in the normal control (NC) and Alzheimer’s disease (AD) groups.

Feature Selection

Feature selection uses specific algorithms to select the most relevant features with the most contribution towards predicting variables for increasing the accuracy and reducing the prediction time. A high-dimensional feature vector, 71 volumetric, and 68×3 surface-based features (cortical thickness, sulcal depth, and gyrification index) were in this study without any significant or appropriate information to diagnose AD. Therefore, the following algorithms are used to obtain the top-ranked features, including the Relief-F algorithm, t-test, and one-way ANOVA.

Relief-F Algorithm

Relief-F is one of the filter methods used for feature selection that is particularly sensitive to feature interactions [ 43 ], designed originally for binary classification, and replaced with Relief-F as the most utilized algorithm [ 44 ].

This algorithm aimed to assess the quality of features according to the ability of their values to separate between the cases that are close to each other [ 45 ], including three important steps: the nearest hit and miss, calculation of the weights of features, and presentation of a ranked list of features. Based on this list of features, the top 12 ranked optimal features were selected. The t-test is a statistical test to determine a difference in the means of two samples and either dependent or independent samples. T-tests were used as a feature ranking algorithm in a variety of machine-learning studies [ 46 , 47 ]. The formula of the t-test is defined as follows:

$\mathrm{t-value}=\frac{{\mu }_1-{\mu }_2}{\sqrt{\frac{{\sigma }_1^2}{n_1}+\frac{{\sigma }_2^2}{n_2}}}$ (1)

where n₁, n₂ are the number of samples, μ₁, μ₂ are the means, and σ₁, σ₂ are the standard deviation of two classes.

T-value measures the significance of the difference between two samples relative to the variation in each sample. Therefore, the high t-value of a specific feature for the two samples AD and NC leads to reliability in the classification and selection.

The absolute t-value for each feature was computed, and all features were ranked depending on their t-values. The 12 top discriminative features were selected.

One Way ANOVA

ANOVA stands for analysis of variance was used to compare the sample means for two independent groups, or more, determining whether one group has a statistically significant difference in its mean than the others based on the following formula:

$\mathrm{F-value}=\frac{{\mathrm{MS}}_b}{{\mathrm{MS}}_w}$ (2)

${\mathrm{MS}}_b=\frac{{\sum }_{\mathrm{i=1}}^k{n}_i{(\underset{1}{\overset{-}{x}}-\stackrel{-}{x})}^2}{k-1}$ (3)

${\mathrm{MS}}_w=\frac{{\sum }_{\mathrm{i=1}}^k{\sum }_{\mathrm{j=1}}^n(x_{\mathrm{ij}}-\underset{1}{\overset{-}{x}})}{n-k}$ (4)

where F is the variance ratio for the overall test, MS_b is the mean square between groups, MS_w is the mean square within groups, k is the number of classes, and n is the number of observations.

The F-value was measured for all features and ranked from the highest F-value to the lowest; the 12 top-ranked features were then obtained.

Classification

Support Vector Machine (SVM)

SVM is a discriminative classifier for the selection of the best hyper-plane or a group of hyper-planes that maximizes the distance of the margin to classify the data into many classes. The hyperplane is defined by the following equation:

$\mathrm{g(x)}=\mathrm{wTx}+b$

where w is the weight vector, and b is the offset parameter for the input vector x.

The maximum and minimum margin widths are 2/(||w||) and 1/2 ||w||, respectively.

For non-linearly separable data, SVM uses a kernel function with an added dimension to the data and transforms data to a higher-dimensional space, such as the Gaussian kernel defined [ 48 ] as follows:

$\mathrm{(x,y)}=\exp \left(\mathrm{-\gamma }{|x-y|}^2\right)$

where γ is gamma, and ${|x-y|}^2$ is defined as squared Euclidean distance between the two feature vectors. The gamma hyperparameter (γ) controls the training points, which affected the decision boundary.

K-Nearest Neighbor

In the training phase, K-Nearest Neighbor, as one of the simplest supervised machine-learning classifiers, stores and arranges all labeled data in the memory. Therefore, it is memory-based without any need to model fitting and classifies the test point based on a similarity measure between the test point and its nearest neighbors. For example, with x0 as a new point, the k-nearest neighbor search obtained the k closest points in distance to x0. Among these k neighbors, the number of the data points in each class was counted. Based on the most votes from the neighbors, the data point is classified.

Weighted KNN takes the majority votes from the neighbors without caring about their distance from the test point.

Decision Tree

The decision tree is a classification model in a shape of a diagram used in data analysis. In the training step, this algorithm aimed to divide the data into smaller sets of data based on a specific feature. The node in the tree states a condition of a feature; each branch falling from that node corresponds to one of the possible attribute values. Each leaf represents class labels related to the case. Cases in the training set are classified by guiding them from the tree’s root down to a leaf, depending on the result of the tests.

Results

A total of 154 individuals participated in this study, 108 and 46 for the training and testing the performance of classifiers, respectively. The features were organized into four main groups: volume features, cortical thickness, sulcal depth, and gyrification index. Volume was measured for 71 regions of interest (ROI) in the brain. Each of the other three features was measured for 68 ROIs.

The t-test, Relief-F algorithm, and ANOVA were used for the feature ranking process and selected the 12 top-ranked features from each of them, as indicated in Table 2. The 9 common features were selected among the 12 top-ranked features (group1), including the right amygdala, cortical thickness left entorhinal, left amygdala, left hippocampus, cortical thickness right entorhinal, right ambient and parahippocampus, right hippocampus, left ambient and parahippocampus, and left inferior middle temporal gyri.

A total of 9 top-ranked features were then selected, and then the 7 common features were selected among the 9 features. The 7 Common features were considered group2, including the right amygdala, cortical thickness left entorhinal, left Amygdala, left hippocampus, right ambient and parahippocampus, right hippocampus, left ambient, and parahippocampus.

The four classifiers, such as decision tree, linear SVM, Gaussian SVM, and weighted KNN were executed using all features combined and the two groups of features with 46 test points to assess the performance of the proposed feature selection, as shown in Table 3.

Linear SVM and weighted KNN classifiers showed the best performance with 100% accuracy when these 7 features were used (Tables 3 and 4, and Figures 3 and 4). Further, Table 3 illustrates that the average time required for all classifiers to predict one observation when using 7 features is much less compared to the prediction time for using the original features. As a result, the suggested approach would provide the most critical characteristics with the least time and the greatest accurate outcomes compared to earlier efforts. We considered the following measurements:

$\mathrm{Sensitivity}=\frac{\mathrm{TP}}{(\mathrm{TP}+\mathrm{FN})},\mathrm{Specificity}=\frac{\mathrm{TN}}{\mathrm{TN}+\mathrm{FP}}$

$\mathrm{Precision}=\frac{\mathrm{TP}}{\mathrm{TP}+\mathrm{FP}},\mathrm{Accuracy}=\frac{\mathrm{TP}+\mathrm{TN}}{\mathrm{TP}+\mathrm{TN}+\mathrm{FP}+\mathrm{FN}}$

where TP, TN, FP, and FN are true positive, true negative, false positive, and false negative, respectively.

Figure 3. Performance measurements for 7 common features

Figure 4. Performance measurements for 9 common features

Figure 5 shows that cortical thickness right entorhinal and left inferior middle temporal gyri features, excluded when creating group 2, have large overlapping areas between AD (class 1) and NC (class 2). Therefore, the elimination of them increased the accuracy of detection for LSVM and W- KNN.

Figure 5. Boxplot of the two excluded features

Discussion

In this work, 4 machine-learning models were proposed, including decision tree, linear SVM, Gaussian SVM, and weighted KNN for differentiating AD individuals from brain MRI images. Based on the results, linear SVM and weighted KNN achieved the same performance with accuracy of 100% using 7 features. The SVM and KNN provide good performance with 7 and 9 features with sensitivity (recall), selected as the models to decrease missed AD cases as much as possible.

When volume features were combined with surface-based features (CT, SD, and GI), sulcal depth and gyrification index were underestimated. As a result, sulcal depth and gyrification index did not rank amongst the top features. GI and sulcal depth (SD) do not contribute to the detection of AD stage compared to volume features. Therefore, we can rely primarily on the volumes of the hippocampus (left and right), amygdala (left and right), and parahippocampus (left and right) as parts of the limbic brain system. Furthermore, the left cortical thickness of the entorhinal cortex can be added to the previous volume features to improve detection performance.

Table 5 compares various results from previous techniques for detecting Alzheimer’s disease and the proposed method. One compared study developed an approach for classifying AD from NC with accuracy up to 92.86% by using fusion of texture and morphemtric features, RFE-SVM for the feature selection process and SVM for the classification process [ 40 ]. Another study depended on 31 morphemtric features selected using RFE algorithm to differentiate between AD and NC with accuracy equal to 89.81% [ 41 ]. One report developed a method based on 12 layers convolutional neural network (CNN) model for AD diagnosis with an accuracy of 97.65% using MRI images acquired from OASIS database [ 49 ].

The main reason for this output is using a small number of very associated features with AD and removing redundant features. The existence of unrelated features reduces the classification ability of the model and the overall accuracy, showing the enhancement in the models’ performance when excluding right cortical thickness of entorhinal and left inferior temporal gyrus features from the 9 features. Furthermore, this study didn’t depend on one feature selection methods to select features.

The limitation of the proposed study is to use filter feature selection technique without consideration of the features correlation or dependency.

The presented work can be improved by using the MCI stage in the future, requiring more relevant features and implementation of more feature engineering steps, which we are working on to develop an approach to classify the three stages of Alzheimer’s NC, MCI, and AD.

Conclusion

In this study, an efficient classification system for Alzheimer’s disease diagnosis is proposed, based on combining more than one feature selection method (t test, one way ANOVA, and Relief-F algorithm) to acquire the most significant features representing AD from a huge pool of features. Furthermore, four classifiers (decision tree, linear SVM, Gaussian SVM, weighted KNN) was applied to select the highest accuracy. The experiment explained that linear SVM and weighted KNN and the following features are the most precise classifiers: left hippocampus, right hippocampus, left amygdala, right amygdala, left ambient and parahippocampus, right ambient and parahippocampus, and cortical thickness_left entorhinal. In addition, combining volume features with cortical thickness features will provide more accurate results than using either volume or cortical thickness independently. However, the traditional techniques of classifiers have been used and applied on extracted features, the maximum accuracy together with the minimum number of features have been collected.

In the future of this study, we plan to implement an approach to classify the three stages of Alzheimer’s NC, MCI, AD and to increase the dataset for a robust classification system.

Acknowledgment

Data used in the preparation of this article was acquired from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) database (adni.loni.usc.edu). As such, the investigators within the ADNI contributed to the design and implementation of ADNI and/or provided data but did not participate in the analysis or writing of this paper. Data collection and sharing for this project were funded by the Alzheimer’s Disease Neuroimaging Initiative (ADNI) (National Institutes of Health Grant U01 AG024904) and DOD ADNI (Department of Defense award number W81XWH-12-2-0012). ADNI is funded by the National Institute on Aging, the National Institute of Biomedical Imaging and Bioengineering, and through generous contributions from the following: AbbVie, Alzheimer’s Association; Alzheimer’s Drug Discovery Foundation; Araclon Biotech; BioClinica, Inc.; Biogen; Bristol-Myers Squibb Company; CereSpir, Inc.; Cogstate; Eisai Inc.; Elan Pharmaceuticals, Inc.; Eli Lilly and Company; EuroImmun; F. Hoffmann-La Roche Ltd and its affiliated company Genentech, Inc.; Fujirebio; GE Healthcare; IXICO Ltd.;Janssen Alzheimer Immunotherapy Research & Development, LLC.; Johnson & Johnson Pharmaceutical Research & Development LLC.; Lumosity; Lundbeck; Merck & Co., Inc.;Meso Scale Diagnostics, LLC.; NeuroRx Research; Neurotrack Technologies; Novartis Pharmaceuticals Corporation; Pfizer Inc.; Piramal Imaging; Servier; Takeda Pharmaceutical Company; and Transition Therapeutics. The Canadian Institutes of Health Research is providing funds to support ADNI clinical sites in Canada. Private sector contributions are facilitated by the Foundation for the National Institutes of Health (www.fnih.org). The grantee organization is the Northern California Institute for Research and Education, and the study is coordinated by the Alzheimer’s Therapeutic Research Institute at the University of Southern California. ADNI data are disseminated by the Laboratory for Neuro Imaging at the University of Southern California.

Authors’ Contribution

EM. Arabi and KS. Ahmed conceived the design of the study. EM. Arabi accomplished the gathering and analyzing of the data, developed and applied the approach of the study, prepared the original draft. KS. Ahmed carried out the critical revision of the manuscript. All the steps of the study were supervised by KS. Ahmed and AS. Mohra All authors have read and agreed to the published version of the manuscript.

Ethical Approval

All data have been taken under the ADNI approval. ADNI protocol and ethics statement: http://adni.loni.usc.edu/wp-content/themes/freshnews-dev-v2/documents/clinical/ADNI-2_Protocol.pdf.

Conflict of Interest

None

References

1. Brookmeyer R, Gray S, Kawas C. Projections of Alzheimer’s disease in the United States and the public health impact of delaying disease onset. Am J Public Health. 1998; 88(9):1337-42. Publisher Full Text | DOI | PubMed
2. Alzheimer’s Association. Why-Get-Checked. 2022 [cited 2022 March 15]. Available from: https://www.alz.org/alzheimers-dementia/diagnosis/why-get-checked.
3. Social Care Institute for Excellence. Dementia: Why early diagnosis of dementia is important. 2022 [cited 2022 March 15]. Available from: https://www.scie.org.uk/dementia/symptoms/diagnosis/early-diagnosis.asp.
4. Arevalo-Rodriguez I, Smailagic N, Roqué I, Figuls M, Ciapponi A, Sanchez-Perez E, et al. Mini-Mental State Examination (MMSE) for the detection of Alzheimer’s disease and other dementias in people with mild cognitive impairment (MCI). Cochrane Database Syst Rev. 2015; 2015(3):CD010783. Publisher Full Text | DOI | PubMed
5. Wen J, Thibeau-Sutre E, Diaz-Melo M, Samper-González J, Routier A, et al. Convolutional neural networks for classification of Alzheimer’s disease: Overview and reproducible evaluation. Med Image Anal. 2020; 63:101694. DOI | PubMed
6. Bartos A, Gregus D, Ibrahim I, Tintěra J. Brain volumes and their ratios in Alzheimer´s disease on magnetic resonance imaging segmented using Freesurfer 6. Psychiatry Res Neuroimaging 2019; 287:70-74. DOI | PubMed
7. Cuingnet R, Gerardin E, Tessieras J, Auzias G, Lehéricy S, et al. Automatic classification of patients with Alzheimer’s disease from structural MRI: a comparison of ten methods using the ADNI database. Neuroimage. 2011; 56(2):766-81. DOI | PubMed
8. Davatzikos C, Bhatt P, Shaw LM, Batmanghelich KN, Trojanowski JQ. Prediction of MCI to AD conversion, via MRI, CSF biomarkers, and pattern classification. Neurobiol Aging. 2011; 32(12):2322.e19-27. Publisher Full Text | DOI | PubMed
9. Braak H, Braak E. Frequency of stages of Alzheimer-related lesions in different age categories. Neurobiol Aging. 1997; 18(4):351-357. DOI | PubMed
10. Schilling LP, Zimmer ER, Shin M, Leuzy A, Pascoal TA, et al. Imaging Alzheimer’s disease pathophysiology with PET. Dement Neuropsychol. 2016; 10(2):79-90. Publisher Full Text | DOI | PubMed
11. Marcus C, Mena E, Subramaniam RM. Brain PET in the diagnosis of Alzheimer’s disease. Clin Nucl Med. 2014; 39(10):e413-26. Publisher Full Text | DOI | PubMed
12. Gray KR, Wolz R, Heckemann RA, Aljabar P, et al. Multi-region analysis of longitudinal FDG-PET for the classification of Alzheimer’s disease. Neuroimage. 2012; 60(1):221-9. Publisher Full Text | DOI | PubMed
13. Nordberg A, Rinne JO, Kadir A, Långström B. The use of PET in Alzheimer disease. Nat Rev Neurol. 2010; 6(2):78-87. DOI | PubMed
14. Ou YN, Xu W, Li JQ, Guo Y, Cui M, Chen KL, et al. FDG-PET as an independent biomarker for Alzheimer’s biological diagnosis: a longitudinal study. Alzheimers Res Ther. 2019; 11(1):57. Publisher Full Text | DOI | PubMed
15. Graña M, Termenon M, Savio A, Gonzalez-Pinto A, Echeveste J, Pérez JM, Besga A. Computer aided diagnosis system for Alzheimer disease using brain diffusion tensor imaging features selected by Pearson’s correlation. Neurosci Lett. 2011; 502(3):225-9. DOI | PubMed
16. Lee W, Park B, Han K. Classification of diffusion tensor images for the early detection of Alzheimer’s disease. Comput Biol Med. 2013; 43(10):1313-20. DOI | PubMed
17. Sørensen L, Igel C, Pai A, Balas I, Anker C, et al. Differential diagnosis of mild cognitive impairment and Alzheimer’s disease using structural MRI cortical thickness, hippocampal shape, hippocampal texture, and volumetry. Neuroimage Clin. 2016; 13:470-82. Publisher Full Text | DOI | PubMed
18. Achterberg HC, Van Der Lijn F, Den Heijer T, Vernooij MW, Ikram MA, et al. Hippocampal shape is predictive for the development of dementia in a normal, elderly population. Hum Brain Mapp. 2014; 35(5):2359-71. Publisher Full Text | DOI | PubMed
19. Madusanka N, Choi HK, So JH, Choi BK. Alzheimer’s Disease Classification Based on Multi-feature Fusion. Curr Med Imaging Rev. 2019; 15(2):161-9. DOI | PubMed
20. Sørensen L, Igel C, Liv Hansen N, Osler M, Lauritzen M, et al. Early detection of Alzheimer’s disease using MRI hippocampal texture. Hum Brain Mapp. 2016; 37(3):1148-61. Publisher Full Text | DOI | PubMed
21. Liu T, Lipnicki DM, Zhu W, Tao D, Zhang C, Cui Y, Jin JS, Sachdev PS, Wen W. Cortical gyrification and sulcal spans in early stage Alzheimer’s disease. PLoS One. 2012; 7(2):e31083. Publisher Full Text | DOI | PubMed
22. Racine AM, Brickhouse M, Wolk DA, et al. The personalized Alzheimer’s disease cortical thickness index predicts likely pathology and clinical progression in mild cognitive impairment. Alzheimers Dement (Amst). 2018; 10:301-10. Publisher Full Text | DOI | PubMed
23. Cai K, Xu H, Guan H, Zhu W, Jiang J, Cui Y, Zhang J, Liu T, Wen W. Identification of Early-Stage Alzheimer’s Disease Using Sulcal Morphology and Other Common Neuroimaging Indices. PLoS One. 2017; 12(1):e0170875. Publisher Full Text | DOI | PubMed
24. Boutet C, Chupin M, Lehéricy S, Marrakchi-Kacem L, Epelbaum S, et al. Detection of volume loss in hippocampal layers in Alzheimer’s disease using 7 T MRI: a feasibility study. Neuroimage Clin. 2014; 5:341-8. Publisher Full Text | DOI | PubMed
25. Braak H, Braak E. Neuropathological stageing of Alzheimer-related changes. Acta Neuropathol. 1991; 82(4):239-59. DOI | PubMed
26. Poulin SP, Dautoff R, Morris JC, et al. Amygdala atrophy is prominent in early Alzheimer’s disease and relates to symptom severity. Psychiatry Res. 2011; 194(1):7-13. Publisher Full Text | DOI | PubMed
27. Prestia A, Boccardi M, Galluzzi S, Cavedo E, Adorni A, et al. Hippocampal and amygdalar volume changes in elderly patients with Alzheimer’s disease and schizophrenia. Psychiatry Res. 2011; 192(2):77-83. DOI | PubMed
28. Weiner MW, Veitch DP, Aisen PS, Beckett LA, Cairns NJ, et al. The Alzheimer’s Disease Neuroimaging Initiative: a review of papers published since its inception. Alzheimers Dement. 2012; 8(1 Suppl):S1-68. Publisher Full Text | DOI | PubMed
29. Devanand DP, Bansal R, Liu J, Hao X, Pradhaban G, Peterson BS. MRI hippocampal and entorhinal cortex mapping in predicting conversion to Alzheimer’s disease. Neuroimage. 2012; 60(3):1622-9. Publisher Full Text | DOI | PubMed
30. Rohini P, Sundar S, Ramakrishnan S. Characterization of Alzheimer conditions in MR images using volumetric and sagittal brainstem texture features. Comput Methods Programs Biomed. 2019; 173:147-55. DOI | PubMed
31. Apostolova LG, Green AE, Babakchanian S, Hwang KS, Chou YY, Toga AW, Thompson PM. Hippocampal atrophy and ventricular enlargement in normal aging, mild cognitive impairment (MCI), and Alzheimer Disease. Alzheimer Dis Assoc Disord. 2012; 26(1):17-27. Publisher Full Text | DOI | PubMed
32. Klöppel S, Stonnington CM, Chu C, Draganski B, Scahill RI, et al. Automatic classification of MR scans in Alzheimer’s disease. Brain. 2008; 131(Pt 3):681-9. Publisher Full Text | DOI | PubMed
33. Toshkhujaev S, Lee KH, Choi KY, Lee JJ, Kwon GR, Gupta Y, Lama RK. Classification of Alzheimer’s Disease and Mild Cognitive Impairment Based on Cortical and Subcortical Features from MRI T1 Brain Images Utilizing Four Different Types of Datasets. J Healthc Eng. 2020; 2020:3743171. Publisher Full Text | DOI | PubMed
34. Zhu Y, Huang C. An improved median filtering algorithm for image noise reduction. Physics Procedia. 2012; 25:609-16. DOI
35. Kruthika KR, Rajeswari H, Maheshappa HD. Multistage classifier-based approach for Alzheimer’s disease prediction and retrieval. Informatics in Medicine Unblocked. 2019; 14:34-42. DOI
36. Acharya UR, Fernandes SL, WeiKoh JE, Ciaccio EJ, et al. Automated Detection of Alzheimer’s Disease Using Brain MRI Images- A Study with Various Feature Extraction Techniques. J Med Syst. 2019; 43(9):302. DOI | PubMed
37. Beheshti I, Demirel H, Matsuda H. 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. 2017; 83:109-19. DOI | PubMed
38. Mahanand BS, Suresh S, Sundararajan N, Kumar M. IEEE: San Jose, CA, USA; 2011.
39. Kumar MA, Mahanand BS. Springer: New Delhi; 2013. DOI
40. Ding Y, Zhang C, Lan T, Qin Z, Zhang X, Wang W. IEEE: Washington, DC, USA; 2015. DOI
41. Mahanand BS, Babu GS, Suresh S, Sundararajan N. IEEE: Noida, India; 2015. DOI
42. Hammers A, Allom R, Koepp MJ, Free SL, Myers R, Lemieux L, Mitchell TN, Brooks DJ, Duncan JS. Three-dimensional maximum probability atlas of the human brain, with particular reference to the temporal lobe. Hum Brain Mapp. 2003; 19(4):224-47. Publisher Full Text | DOI | PubMed
43. Kira K, Rendell LA. AAAI Press: San Jose, California; 1992.
44. Kononenko I, Bergadano F, Luc D. Springer: Berlin, Heidelberg; 1994. DOI
45. Zhou X, Wang J. Feature Selection for Image Classification Based on a New Ranking Criterion. Journal of Computer and Communications. 2015; 3(3):74-9. DOI
46. Liu M, Zhang D, Shen D. Ensemble sparse classification of Alzheimer’s disease. Neuroimage. 2012; 60(2):1106-16. Publisher Full Text | DOI | PubMed
47. Chaves R, Ramírez J, Górriz JM, López M, Salas-Gonzalez D, Alvarez I, Segovia F. SVM-based computer-aided diagnosis of the Alzheimer’s disease using t-test NMSE feature selection with feature correlation weighting. Neurosci Lett. 2009; 461(3):293-7. DOI | PubMed
48. Keerthi SS, Lin CJ. Asymptotic behaviors of support vector machines with Gaussian kernel. Neural Comput. 2003; 15(7):1667-89. DOI | PubMed
49. Hussain E, Hasan M, Hassan SZ, Azmi TH, Rahman MA, Parvez MZ. IEEE Xplore: Kristiansand, Norway; 2020. DOI
50. Li A, Li F, Elahifasaee F, Liu M, Zhang L. Hippocampal shape and asymmetry analysis by cascaded convolutional neural networks for Alzheimer’s disease diagnosis. Brain Imaging Behav. 2021; 15(5):2330-9. DOI | PubMed
51. Mahanand BS, Suresh S, Sundararajan N, Kumar M. IEEE: Kharagpur, India; 2013. DOI