Mobile QR Code

1. (Dept. of Information and Communication Engineering, Chosun University, 309 Pilmun-Daero, Dong-Gu, Gwangju 61452, Korea)

sMRI, Alzheimer’s disease, SVM-RFE, SVM, NB, KNN, Mild cognitive impairment, CSF, Genetics, Hippocampus

## 1. Introduction

Alzheimer’s disease (AD), which mostly affects the elderly, is a common neurodegenerative brain disease. Mild cognitive impairment (MCI) is a stage in which a person has a mild, but noticeable, change in thinking patterns. Although there is no medication to cure Alzheimer’s, some medications have been prescribed to delay the onset of memory-related symptoms in patients [1]. Patients with MCI have a high risk of progressing to dementia [2,3]. Some MCI patients progress to AD after baseline within a certain time frame, while others remain stable. Reports have shown that 10% to 15% of MCI patients’ progress to AD per year, and 80% of them will convert to AD after five to six years of follow-up [3]. It is crucial to find biomarkers that can classify patients who have MCI and who will later progress to AD (converter MCI) from those who convert to AD and from those who remain a healthy control (HC).

The most practiced principles for clinical analysis of AD were developed and established about 30 years back by the National Institute of Neurological, Communicative Disorders and Stroke—Alzheimer’s Disease and Related Disorders Association (NINCDS-ADRDA) [4]. However, those principles have been proclaimed inaccurate in up to 20% of cases when practiced in specialized research institutes on patients under later-phase observation over several years [5], and they have provided specificity and sensitivity ranging from 44.3% to 70.8% and from 70.9% to 87.3%, respectively [6]. Therefore, the benchmark may lead to even more inaccurate diagnoses in AD patients in earlier stages of the disease, surprisingly for those with MCI. Due to this, there has been a serious need to improvise in order to improve the efficiency of diagnosis. It was believed that imaging and biological markers could produce this enhanced accuracy [7]. Consequently, two recent revisions of the NINCDS-ADRDA criteria took place, one by the Alzheimer’s Association and National Institute of Aging (NIA) [2] and the other by Dubois et al.[8]. Both improvised criteria now recommend the use of imaging and non-imaging biomarkers to support a proper diagnosis of both AD and MCI. However, only five largely studied biomarkers of AD were integrated into the medication criteria. The former modification demonstrated that biomarkers are meant to complement the preeminent clinical treatment, and the latter highly recommended their addition to improve AD investigations, despite being only used in institutional research. Biomarkers being taken into consideration are low levels of the 42-amino-acid alteration of A$\beta$ (A$\beta$42) in cerebrospinal fluid (CSF) and elevated CSF phosphorylated tau (P-tau) or total tau (T-tau). CSF biomarkers that have been applied in several studies in the literature include P-tau, T-tau, and A$\beta$42. These three CSF biomarkers provide valuable information for AD identification, because patients have drastically low levels of A$\beta$42 and high levels of P-tau and T-tau [9]. It has been determined that the combination of CSF and T-tau markers provides outstanding classification efficiency for separating AD patients from HC, with high specificity and sensitivity [10]. Besides these biomarkers, other interesting biomarkers for AD detection include the presence of the APoE ε4 allele in patients. For large numbers of APoE ε4 alleles, CSF studies [11] have been carried out for the diagnosis of AD and MCI. Moreover, genetic risk factors also play a vital role in imaging and biological markers for AD diagnosis. One previous study in the literature [12] proved that the presence of a specific variant of the apolipoprotein E gene (APOE) is a major risk factor related to late-onset AD. APOE has three major alleles: ε2, ε3, and ε4. In observations, AD patient carriers of the ε4 allele generally have low CSF A$\beta$42 and high levels of P-tau and T-tau, along with rapid atrophy patterns on MRI. Diverse aspects of pathological patterns linked to AD can be revealed by different biomarkers; therefore, independent biomarkers might assist in proper diagnosis. It has been shown that multimodal biomarkers can enhance diagnostic accuracy [13-15]. A recent study of note [16] demonstrated the excellence of machine learning approaches. One of the widely utilized methods for solving the classification task is the support vector machine (SVM). Several studies have applied the SVM to AD identification and classification [14].

Hippocampal atrophy, cortical thickness, genetics, and CSF composition changes are considered to be the major hallmarks of AD, and are therefore used as diagnosis markers [17]. Reduced hippocampal volume shows a strong association with the Alzheimer’s disease [18] pattern of hippocampal atrophy, and can be precisely utilized to identify AD, which plays a vital role in the clinical detection of AD [19]. Moreover, alteration of cortical thickness [20], as well as a decrease in hippocampal volume and an alteration of CSF composition have been demonstrated in patients with AD, in comparison to healthy controls [9]. In this paper, we propose a classification framework to precisely diagnose individuals with Alzheimer’s disease and mild cognitive impairment from healthy (normal) controls. First, we utilized the FreeSurfer pipeline to separately obtain cortical thickness and hippocampal volume [21]. After that, we combined all these measures into a predictive model, and calculated the performance from classification. We hypothesize that feature combination will outperform the separate, individual model. Fig. 1 represents the workflow of the proposed method; the rest of the flow in the proposed method is as follows.

## 2. Material and Method

### 2.1 Data

All data for the individuals used in this analysis were collected from the Alzheimer’s Disease Neuroimaging Initiative (ADNI). The ADNI was initiated in 2003 as a public-private partnership under the supervision of Michael W. Weiner, MD. The primary aim of ADNI has been to test whether positron emission tomography, magnetic resonance imaging, other clinical and biological markers, and neuropsychological test assessments can be combined for early Alzheimer’s disease prediction and for MCI progression. Demographic information, raw neuroimaging data, CSF measures, APOE genotypes, diagnostic information, and neuropsychological test scores are publicly available from the ADNI data repository (http://adni.loni.usc.edu). In this paper, a total of 217 subjects were used, including 53 AD patients, 103 MCI patients, and 61 healthy controls. Table 1 presents the demographics of all these subjects. All structural MRI scans for this paper were obtained from 1.5 T scanners.

The entries for age, gender, education, and MMSE denote mean and standard deviation for each group. MMSE, Mini-Mental State Exam; CDR, clinical dementia ratio.

##### Table 1. Demographics of the participants.
 Group HC MCI AD No. of Subjects 61 103 53 Male/Female 28/33 68/35 32/21 Age 75.3 ± 5.2 75.3 ± 7.0 75.2 ± 7.4 MMSE 29 ± 1.2 27.1 ± 1.7 23.8 ± 2.0 CDR 0 0.5 0.7 ± 0.3 Education 15.8 ± 3.2 15.9 ± 2.9 14.7 ± 3.6

### 2.2 Data Acquisition

We downloaded all the MRI images in Neuroimaging Informatics Technology Initiative (NifTi) format. Downloaded images were preprocessed for spatial distortion and B1 field inhomogeneity correction.

CSF data were collected in the morning after overnight fasting with a 20-G or 24-G spinal needle. Within one hour of acquisition, CSF biomarkers were frozen and transported to the ADNI core laboratory at the Medical Center of the University of Pennsylvania.

The ADNI biomarker core laboratory also provided genotype and gene expression data for each participant in this study, which were obtained from peripheral blood samples. The genetic feature was a single categorical variable for each participant, taking one of five possible values: (ε2, ε3), (ε2, ε4), (ε3, ε3), (ε3, ε4), or (ε4, ε4). In this study, we specifically analyzed APoE ε4 allele status (carrier vs. non-carrier).

### 2.3 FreeSurfer Analysis of MRI

We applied the recon-all FreeSurfer pipeline, version 6.0.0 (http://surfer.nmr.mgh.harvard.edu) to the MRI images for cortical reconstruction and volumetric segmentation [21]. This pipeline automatically generates reliable volume and thickness segmentation of white matter and gray matter, as well as subcortical volume. Subcortical volumetric segmentation and cortical reconstruction included removal of non-brain parts, Talairach-transformations, segmentation of subcortical gray matter and white matter regions, intensity standardization, and atlas registration. After these steps, a cortical surface mesh model was generated, and finally, the 34 cortical regions were obtained from cortical surface parcellation based on sulcal and gyral landmarks for both hemispheres corresponding to the Desikan-Killiany atlas [22].

### 2.4 Hippocampal Volume

Hippocampal segmentation was performed using the FreeSurfer [23] tool. Hippocampal volume is considered one of the major hallmarks in Alzheimer’s disease identification [24] due to the fact that detailed analysis of the hippocampus is considered a major step in the analysis. Thus, the hippocampus has been one of the frequently studied structures for diagnosing AD. However, this structure is not homogeneous, so it is usually subdivided into different subfields. Initial efforts to define the hippocampus subfields were mainly based on cell size, shape, and connectivity [25]. FreeSurfer separately segments the hippocampus subfields into right and left parts, which gives an estimate of the probability that every individual voxel associated with a certain arrangement is based on a priori insight regarding spatial relationships, which are obtained by using a training set. Differences in voxel intensity are located, and the subcortical region is parcellated, and then, affine registration in the Talairach space is performed. The detailed FreeSurfer processing stages are presented in [21], and the hippocampus subfield segmented regions are shown in Fig. 2.

### 2.5 Cortical Thickness

To calculate cortical thickness, T1‐weighted images were preprocessed using FreeSurfer [21]. To create high-contrast-to-noise-ratio images from brain-extracted images, intensity normalization was applied. The gray matter and white matter boundary was located by using images obtained after intensity normalization. After this, a triangular mesh around the white matter surface was constructed. Each brain hemisphere was then fragmented over 160,000 vertices of the triangular mesh. The mesh was outwardly deformed so the grey matter surface was created, and the boundary between the cerebral spinal fluid and grey matter surface followed the boundary. Cortical thickness was measured as the distance between the grey matter surface and the white matter surface for each vertex. A FreeSurfer common template was used to register the image using the cortical folding pattern of images. The neocortex was then parcellated into 68 neocortical brain regions (34 brain regions for each hemisphere) based on the Desikan-Killiany atlas as shown in Fig. 3. [22]. Mean thickness within that parcellation of all the vertices gives the thickness of each parcellation unit. Finally, 68 cortical thickness features were yielded per subject.

### 2.6 Feature Selection

In most studies involving neuroimaging analysis, the number of predictor voxels obtained will outnumber the subjects. Thus, a dimensionality reduction technique is necessary in order to obtain the most relevant feature set, to discard noise and redundant features, and to escape numerical singularities and overfitting issues, thus enhancing classification efficiency. An efficient feature reduction algorithm is the essential section of a machine learning technique in cases of high-dimensional feature sets. We have shown the efficiency of the support vector machine-recursive feature elimination (SVM-RFE) algorithm in recognizing the early moments of AD [25]. Importantly, feature reduction was implemented only for the training data. Once identified, the same brain regions used during training were utilized to assess the predictive accuracy of the classifier on the test data. In this study, SVM-RFE was applied in order to obtain a ranked list of features that could best distinguish HC from AD and MCI. SVM-RFE is a multivariate wrapper technique-based feature selection algorithm, which precisely fits a model and eliminates the weakest feature until the defined informative number of features is reached. The SVM-RFE ranking criterion is closely similar to the SVM model. An SVM model is trained in every iteration of RFE. Then, the features with lower ranking criteria are eliminated, since they have the least effect on the classification process, while the remaining feature vectors are kept for the next iteration. This sequence is repeated until all the features have been removed. Then, in ranked order of elimination, the features are graded. A detailed description of the SVM-RFE model can be found in the literature. In this work, after the application of SVM-RFE, the highest ranked training features that maximize cross-validation accuracy were kept for training the classifiers.

### 2.7 Classification

The basic classification protocol for the proposed prediction framework is shown in Fig. 1. The machine learning framework consists of four major steps: feature extraction, feature selection, normalization, and classification. Brief descriptions of each of the classifiers used in this experiment are described here.

#### 2.7.1 NB

NB is a machine-machine learning technique that has been practiced for more than 50 years in the field of biomedical informatics. This classifier is a probabilistic machine learning model that is used for the classification task. The crux of the classifier is based on Bayes’ theorem with a strong independence assumption among the features. It has an easy construction model, with no complex iterative framework approximations, which makes it particularly effective for large datasets. Despite its simplicity, it performs unexpectedly well for classification tasks, and is widely utilized because it usually outperforms several other sophisticated techniques. Bayes’ theorem determines posterior probability p(x/y) from p(x), p(y), and p(y/x). The NB classifier estimates that the effect of the criteria of the predictor (y) on a specified class (x) is an autonomous process, compared to another predictor’s criteria. This assumption is known as class-conditional independence:

##### (1)
$p(x / y)=\frac{p(y / x) p(x)}{p(y)}$

where

$$p(x / y)=p\left(y_{1}, / x\right) \times p\left(y_{2} / x\right) \times \ldots \times p\left(y_{n} / x\right) \times p(y) \cdot p(y / x)$$

is the likelihood of the class, given the predictor, in which p(y) is the prior probability of the class, p(x/y) is the predictor probability given the class, and p(y) is the prior probability of the predictor.

#### 2.7.2 KNN

The KNN classifier was proposed by Cover and Hart in 1980. It is a well-known (and the simplest) machine learning classifier. A labeled database is given for training, and then unknown data samples are classified based on the labels of the $\textit{k}$ nearest neighbors. Here, $\textit{k}$ is the major parameter for the KNN algorithm. In this method, the distances between testing data sample $\textit{x}$ and training data sample $x_{i}, I=1, \ldots n$ are calculated:

##### (2)
$d_{E}\left(x, x_{i}\right)=\sum_{i=1}^{n}\left|x-x_{i}\right|$
##### (3)
$x: d_{E}\left(x, x_{i}\right)<d_{E}\left(x, x_{i}\right), i \neq j$

The nearest $\textit{k}$ points are calculated. Testing samples are classified with respect to specified $\textit{k}$ nearest neighbors.

#### 2.7.3 SVM

SVM is a supervised learning algorithm, and is one of the most widely known classification algorithms; its usage has been valuable in a large number of applications, including the classification and prediction of disease from structural MRIs of the human brain. Fig. 4 presents an SVM classifier that classifies data by constructing a hyperplane, which is defined as $w^{T} . x+b=0$, of a very high-dimensional feature vector, where $\textit{b}$ is the bias for the input vector, and $\textit{w}$ is the weight vector. A good identification is said to be the nearest training data sample belonging to any class—the greater the separation distance of the margin, the lower the generalization error of the classifier. Let us assume $\textit{N}$ training data samples $\left\{\left(x_{1}, y_{1}\right),\left(x_{2}, y_{2}\right), \ldots,\left(x_{N}, y_{N}\right)\right\}$ are given; $x_{i} \in \mathbb{R}^{d}$ is a set of feature vectors, and $y_{i} \epsilon\{-1,1\}$ is the class label. Then, the classification problem can be described by the minimization problem below:

##### (4)
$\min _{w, b}\left(\frac{w^{T} w}{2}\right)$
##### (5)
$\text { s.t. : } y_{i}\left(w^{T} . x_{i}\right)+b>1, i=1, \ldots n$

For a small dataset, an RBF kernel performs better than a linear kernel. A regularization constant, $\textit{C}$, and a set of kernel hyperparameters, $\textit{γ}$(gamma), need to be tuned in SVMs. Parameter optimization is obtained by using a cross-validation (CV) technique. This procedure is repeated 1500 times, each time randomly selecting a new set of 10 held-out subjects to obtain the optimum hyperparameter. In this method, the search scales for regularization constants and gamma values were set to C= [0.001, 0.01, 0.1, 1, 10, 100] and $\textit{γ}$= [0.001, 0.01, 0.1, 1, 10, 100], respectively. The maximum validation accuracy was obtained at $\textit{C}$=0.1 and $\textit{γ}$=0.01. The tuned parameters are used to predict the accuracy value on the test dataset.

## 3. Results and Discussion

Performance evaluations are measured in terms of accuracy, sensitivity, specificity, precision, F1 score, and area under the receiver operating characteristic (ROC) curve (AUC) for classification of AD vs. HC, AD vs. MCI, and HC vs. MCI using the four biomarker measures individually, and by combining all four features. The accompanying AUC represents the measure of classification accuracy, as shown in Tables 2-4. Fig. 5 shows the set of the optimal number of features for the all-features combination and Fig. 6 presents the AUC for the combination of all features, as well as a comparison between different classifiers. First, we performed feature selection using SVM-RFE, which is a wrapper method of feature selection. Two reference measures performed reasonably well, especially the hippocampus volume, genetics, and CSF measures, which can discriminate between AD patients and healthy controls quite well, compared to cortical features alone. Tables 2 and 4 show that the hippocampus feature gives the best accuracy for classifying the disease, compared to the other two features. In HC vs. MCI classification, CSF plus genetic measures outperformed the other two features. From the results, we can say that hippocampus volume performs well in the individual features. Most importantly, however, the combination of all four measures outperformed the separate measures. From the results shown in Tables 2-4, we can say that the proposed method achieved a performance accuracy of 96.93%, with sensitivity at 99.89%, specificity at 95.40%, precision at 93.83%, an F1 score of 96.10%, Cohen’s Kappa at 0.9320, and AUC at 98.30% for classifying AD vs. HC. For classifying HC vs. MCI, our method achieved accuracy of 92.20%, with sensitivity at 91.40%, specificity at 86.14%, precision at 87.32%, an F1 score of 85.60%, Cohen’s Kappa at 0.8197, and AUC at 94.90%. Similarly, for classifying AD vs. MCI, our method achieved accuracy of 91.32%, sensitivity at 93.12%, specificity at 89.80%, precision at 92.85%, an F1 score of 90.01%, Cohen’s Kappa at 0.8301, and AUC at 93.23%. From the table, we can also see that SVM classifiers performed better, in comparison to other classifiers, when combining all features, but for individual features, the other classifiers also performed well. For cortical thickness, KNN outperformed other classifiers in AD vs. HC classification. To evaluate the performance of our method, we used SVM-RFE to select the important features, which increased classification accuracy. SVM (RBF) was used to evaluate the proposed algorithm for classification. Moreover, we used nested 10-fold cross validation for hyperparameter optimization for the SVM classifiers, which selects the best parameter and model to give the maximum classification accuracy.

##### Table 2. AD vs. HC classification results.
 AD vs. HC Classifiers AUC ACC SEN SPEC PRE F1 Cohen CTH NB 88.25 81.47 82.51 88.54 83.68 87.30 0.8705 KNN 93.65 85.53 86.81 92.63 88.50 90.63 0.8107 SVM 88.44 86.81 84.81 83.30 84.71 88.01 0.7909 Hippo NB 93.45 85.21 91.50 93.20 96.10 91.78 0.9041 KNN 94.51 83.45 93.95 100 93.53 95.12 0.9084 SVM 96.02 87.22 90.45 92.30 90.77 93.85 0.8920 CSF+ Genetics NB 88.90 80.03 93.40 77.81 88.33 91.15 0.8321 KNN 91.13 82.95 94.20 89.71 92.54 91.10 0.8324 SVM 87.85 84.77 88.95 78.88 89.53 90.75 0.9130 Proposed Method NB 92.45 92.33 94.75 90.93 95.14 93.73 0.8580 KNN 96.93 94.50 97.21 92.63 94.77 96.02 0.9307 SVM 98.30 96.93 99.89 95.40 93.83 96.10 0.9320
##### Table 3. HC vs. MCI classification results.
 HC vs. MCI Classifiers AUC ACC SEN SPEC PRE F1 Cohen CTH NB 87.75 80.33 93.24 88.05 82.99 85.13 0.81.45 KNN 90.85 77.65 90.52 85.13 82.71 81.95 0.8401 SVM 86.58 81.25 88.95 83.83 82.96 87.21 0.8095 Hippo NB 89.75 77.88 91.85 85.44 87.45 81.55 0.8650 KNN 88.45 80.65 90.87 86.75 89.23 84.12 0.8709 SVM 90.47 78.50 85.63 87.90 90.40 87.98 0.8805 CSF+ Genetics NB 89.90 81.66 88.65 90.77 78.45 81.85 0.8212 KNN 90.54 85.97 91.75 87.56 85.41 80.55 0.7545 SVM 88.90 83.70 87.99 78.45 81.20 79.95 0.8057 Proposed Method NB 89.75 85.60 90.33 87.55 85.77 82.85 0.7801 KNN 91.70 89.44 92.87 82.30 87.85 86.33 0.8512 SVM 94.90 92.20 91.40 86.14 87.32 85.60 0.8197
##### Table 4. AD vs. MCI classification results.
 AD vs. MCI Classifiers AUC ACC SEN SPEC PRE F1 Cohen CTH NB 87.95 80.55 91.54 88.67 87.545 83.78 0.8805 KNN 84.30 78.85 89.97 85.63 90.15 85.30 0.8795 SVM 88.55 81.57 93.88 87.75 85.90 83.44 0.8177 Hippo NB 91.65 83.12 95.495 90.12 87.65 90.75 0.87.85 KNN 86.78 84.43 90.01 84.33 86.15 82.90 0.89.88 SVM 84.77 85.31 88.75 83.98 81.85 86.31 0.8802 CSF+ Genetics NB 89.77 81.89 93.01 87.99 84.33 81.95 0.8310 KNN 96.45 78.12 96.10 90.14 88.52 90.55 0.8795 SVM 87.95 82.78 87.20 85.52 88.45 85.97 0.84.33 Proposed Method NB 91.57 87.54 93.90 86.28 92.82 81.52 0.7890 KNN 88.35 85.72 88.83 84.48 83.30 87.23 0.8350 SVM 93.23 91.32 93.12 89.80 92.85 90.01 0.8301

### 3.1 Classification Result

In this method, we utilized a binary classification technique to measure classification performance on the obtained cortical thickness, hippocampus volume, genetics, and CSF biomarkers.

Fig. 6 shows a comparison of AUC for AD vs. HC, HC vs. MCI, and AD vs. MCI. From the above ROC curve and Tables 2-4, we can see that the all-features combination with SVM classifiers outperformed the other classifiers, but for individual features, NB and KNN classifiers also performed better, compared to the SVM. For cortical thickness and CSF plus genetic measures in AD vs. HC classification, the KNN classifier had better AUC compared to the SVM and NB. Similarly, for hippocampus features in classifying AD vs. MCI, NB outperformed the other two classifiers in terms of AUC (91.65%).

### 3.2 Comparison with Other Methods

Recently, several research articles have analyzed neuroimaging and machine learning methods for Alzheimer’s disease diagnosis, with the focus on multimodality techniques. Table 5 shows a comparison of the classification performance by the proposed technique against recently published state-of-the-art methods that used multimodality datasets to diagnose Alzheimer’s disease. However, it is hard to compare the existing state-of-the-art methods, because the majority of the articles used various datasets and validation techniques, both of which influence classification performance. The study by Westman et al. [26] obtained accuracy of 91.8% with sensitivity at 88.5% and specificity at 94.6% for AD vs. healthy group classification of 369 subjects using the orthogonal partial least squares to latent structures (OPLS) method. Hinrichs et al. [27] obtained classification accuracy of 92.40%, with 86.70% sensitivity, 96.60% specificity, and 97.70% AUC when classifying AD vs. healthy patients for 230 subjects by using the MKL technique on multimodal ADNI datasets (i.e. sMRI, CSF, APoE ε4, and cognitive scores). In another study, Zhang and Shen et al. [28] using SVM classifiers obtained accuracy of 93.3% when classifying AD vs. healthy patients from among 186 subjects. Similarly, Liu et al., using multimodal cascaded CNN methods for sMRI and PET images [29], obtained accuracy of 93.26%, sensitivity at 92.55%, and specificity at 93.94%, with 95.68% AUC when classifying 397 subjects. Based on the existing literature, the proposed system (as results show in Table 2) was highly competitive in terms of classification performance, with accuracy of 96.93%, sensitivity at 99.89%, specificity at 95.40%, and AUC at 98.30% for AD vs. healthy patients.

##### Table 5. Classification comparison of the proposed method with existing methods.
 Method Modality Subjects Classifier ACC SEN SPEC AUC Westman et al. [26] sMRI+CSF 369 OPLS 91.80 88.50 94.60 - Hinrichs et al. [27] MRI+PET+CSF+APoE ε4 +cognitive scores 230 MKL 92.40 86.70 96.60 97.70 Zhang and Shen et al. [28] sMRI+PET+CSF 186 SVM 93.30 - - - Liu et al. [29] sMRI+FDG+PET 397 CNN 93.26 92.55 93.94 95.68 Proposed Method sMRI (cortical thickness+hippocampal volume+CSF+APoE ε4) 217 SVM(RBF) 96.93 99.89 95.40 98.30

## 4. Conclusion

In this paper, we first extracted hippocampus volume and cortical thickness using the FreeSurfer toolbox, and obtained CSF and genetics measures from the ADNI database by matching individual subjects. From the results, we can say that hippocampus volume outperforms the other two features (namely, cortical thickness and CSF plus genetics measures). So hippocampus volume shows good accuracy from among the individual features. By combining all four features for AD diagnosis, we noticed that a combination of all features outperformed the individual features. Besides, we used SVM-RFE wrapper feature selection to obtain the optimal feature set, which increased the classification accuracy. Finally, we fed the selected features into an SVM radial bias kernel with 10-fold nested cross validation to obtain the classification results, which demonstrated the effectiveness of the proposed feature selection/combination method to improve classification performance. We used nested CV for hyper-parameter optimization, so we can select the best model for better accuracy. Furthermore, in order to enhance the effectiveness of the proposed method, we plan to increase the number of datasets, include a longitudinal dataset, adding a multimodal dataset and different imaging techniques, such as PET and fMRI, plus different classifiers and other feature selection methods.

## Conflicts of Interest

The authors declare they have no conflicts of interest regarding the publication of this paper.

### ACKNOWLEDGMENTS

This work was supported by the National Research Foundation of Korea (NRF) grant funded by the Korea government (MSIT) (No. NRF-2019R1A4A1029769, NRF-2019R1F1A1060166).

### REFERENCES

1
Prince M., Albanese E., Guerchet M., Prina M., Sep. 2014, World Alzheimer Report 2014: Dementia and risk reduction: An analysis of protective and modifiable risk factors, Accessed: Apr. 03, 2019.
2
Albert M. S., et al. , May 2011, The diagnosis of mild cognitive impairment 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. J. Alzheimers Assoc., Vol. 7, No. 3, pp. 270-279
3
Petersen R. C., Smith G. E., Waring S. C., Ivnik R. J., Tangalos E. G., Kokmen E., Mar. 1999, Mild Cognitive Impairment: Clinical Characterization and Outcome, Arch. Neurol., Vol. 56, No. 3, pp. 303-308
4
McKhann G., Drachman D., Folstein M., Katzman R., Price D., Stadlan E. M., Jul. 1984, Clinical diagnosis of Alzheimer’s disease: report of the NINCDS-ADRDA Work Group under the auspices of Department of Health and Human Services Task Force on Alzheimer’s Disease, Neurology, Vol. 34, No. 7, pp. 939-944
5
Blennow K., Hampel H., Oct. 2003, CSF markers for incipient Alzheimer’s disease, Lancet Neurol., Vol. 2, No. 10, pp. 605-613
6
Beach T. G., Monsell S. E., Phillips L. E., Kukull W., Apr. 2012, Accuracy of the Clinical Diagnosis of Alzheimer Disease at National Institute on Aging Alzheimer Disease Centers, 2005-2010, J. Neuropathol. Exp. Neurol., Vol. 71, No. 4, pp. 266-273
7
Jack C. R., et al. , 2008, The Alzheimer’s disease neuroimaging initiative (ADNI): MRI methods, J. Magn. Reson. Imaging, Vol. 27, No. 4, pp. 685-691
8
Aug. 2007, Research criteria for the diagnosis of Alzheimer’s disease: revising the NINCDS-ADRDA criteria, Lancet Neurol., Vol. 6, No. 8, pp. 734-746
9
Langbaum J. B., et al. , Jul. 2013, Ushering in the study and treatment of preclinical Alzheimer disease, Nat. Rev. Neurol., Vol. 9, No. 7, pp. 371-381
10
Hampel H., Bürger K., Teipel S. J., Bokde A. L. W., Zetterberg H., Blennow K., Jan. 2008, Core candidate neurochemical and imaging biomarkers of Alzheimer’s disease, Alzheimers Dement. J. Alzheimers Assoc., Vol. 4, No. 1, pp. 38-48
11
Mosconi L., et al. , May 2009, FDG-PET changes in brain glucose metabolism from normal cognition to pathologically verified Alzheimer’s disease, Eur. J. Nucl. Med. Mol. Imaging, Vol. 36, No. 5, pp. 811-822
12
Vounou M., et al. , Mar. 2012, Sparse reduced-rank regression detects genetic associations with voxel-wise longitudinal phenotypes in Alzheimer’s disease, NeuroImage, Vol. 60, No. 1, pp. 700-716
13
Zhang D., Wang Y., Zhou L., Yuan H., Shen D., Alzheimer’s Disease Neuroimaging Initiative , Apr. 2011, Multimodal classification of Alzheimer’s disease and mild cognitive impairment, NeuroImage, Vol. 55, No. 3, pp. 856-867
14
Zhang D., Shen D., Alzheimer’s Disease Neuroimaging Initiative , Jan. 2012, Multi-modal multi-task learning for joint prediction of multiple regression and classification variables in Alzheimer’s disease, NeuroImage, Vol. 59, No. 2, pp. 895-907
15
Kohannim O., et al. , Aug. 2010, Boosting power for clinical trials using classifiers based on multiple biomarkers, Neurobiol. Aging, Vol. 31, No. 8, pp. 1429-1442
16
Davatzikos C., Fan Y., Wu X., Shen D., Resnick S. M., Apr 2008, Detection of prodromal Alzheimer’s disease via pattern classification of magnetic resonance imaging, Neurobiol. Aging, Vol. 29, No. 4, pp. 514-523
17
2016, Biomarkers in Alzheimer’s Disease., Elsevier
18
Jack C. R., et al. , Apr. 1999, Prediction of AD with MRI-based hippocampal volume in mild cognitive impairment, Neurology, Vol. 52, No. 7, pp. 1397-1403
19
Feng F., et al. , Sep. 2018, Radiomic Features of Hippocampal Subregions in Alzheimer’s Disease and Amnestic Mild Cognitive Impairment, Front. Aging Neurosci., Vol. 10
20
Lerch J. P., Pruessner J. C., Zijdenbos A., Hampel H., Teipel S. J., Evans A. C., Jul. 2005, Focal decline of cortical thickness in Alzheimer’s disease identified by computational neuroanatomy, Cereb. Cortex N. Y. N 1991, Vol. 15, No. 7, pp. 995-1001
21
Fischl B., Aug. 2012, FreeSurfer, NeuroImage, Vol. 62, No. 2, pp. 774-781
22
Desikan R. S., et al. , Jul. 2006, An automated labeling system for subdividing the human cerebral cortex on MRI scans into gyral based regions of interest, NeuroImage, Vol. 31, No. 3, pp. 968-980
23
Guyon I., Weston J., Barnhill S., Vapnik V., Jan. 2002, Gene Selection for Cancer Classification using Support Vector Machines, Mach. Learn., Vol. 46, No. 1, pp. 389-422
24
Warner H. R., Toronto A. F., Veasey L. G., Stephenson R., Jul. 1961, A Mathematical Approach to Medical Diagnosis: Application to Congenital Heart Disease, JAMA, Vol. 177, No. 3, pp. 177-183
25
Tufail A. B., Abidi A., Siddiqui A. M., Younis M. S., Dec. 2012, Automatic Classification of Initial Categories of Alzheimer’s Disease from Structural MRI Phase Images: A Comparison of PSVM, KNN and ANN Methods, Int. J. Biomed. Biol. Eng., Vol. 6, No. 12, pp. 713-717
26
Westman E., Muehlboeck J.-S., Simmons A., Combining MRI and CSF measures for classification of Alzheimer’s disease and prediction of mild cognitive impairment conversion., NeuroImage. 2012, Vol. 62, No. 1, pp. 229-238
27
Hinrichs C., Singh V., Xu G., Johnson S. C., 2011, Predictive markers for AD in a multi-modality framework: an analysis of MCI progression in the ADNI population., NeuroImage, Vol. 55, No. 2, pp. 574-589
28
Zhang D., Shen D., 2012, Multi-modal multi-task learning for joint prediction of multiple regression and classification variables in Alzheimer’s disease., NeuroImage., Vol. 59, No. 2, pp. 895-907
29
Liu M, Cheng D, Wang K, Wang Y; Alzheimer’s Disease Neuroimaging Initiative., 2018 Oct., Multi-Modality Cascaded Convolutional Neural Networks for Alzheimer's Disease Diagnosis., Neuroinformatics, Vol. 16, No. 3-4, pp. 295-308

## Author

##### Uttam Khatri

Uttam Khatri received his B.Eng. in Electronics and Communication Engi-neering from Pokhara University (Nepal Engineering College), Nepal, in 2015. In 2015-2018, he worked as a junior professor at Nepal Engineering College. Currently, he is a research scholar at Chosun University, Gwang-ju City, Republic of Korea. His research interests include Artificial Neural Networks, Artificial Intelligence Systems, and Machine Learning on Image Processing, especially in Medical Image Processing.

##### Goo-Rak Kwon

Goo-Rak Kwon received an MSc from the School of Electrical and Computer Engineering, SungKyun-Kwan University, in 1999, and a PhD from the Department of Mechatronic Engineering, Korea University, in 2007. He has been a Professor with Chosun University, since 2017. His research interests include medical image analysis, A/V signal processing, video communication, and applications. verification of communication protocols. He is a member of the IEEE.