Three-Dimensional Eigenbrain for the Detection of Subjects and Brain Regions Related with Alzheimer’s Disease

Abstract

Background:

Considering that Alzheimer’s disease (AD) is untreatable, early diagnosis of AD from the healthy elderly controls (HC) is pivotal. However, computer-aided diagnosis (CAD) systems were not widely used due to its poor performance.

Objective:

Inspired from the eigenface approach for face recognition problems, we proposed an eigenbrain to detect AD brains. Eigenface is only for 2D image processing and is not suitable for volumetric image processing since faces are usually obtained as 2D images.

Methods:

We extended the eigenbrain to 3D. This 3D eigenbrain (3D-EB) inherits the fundamental strategies in either eigenface or 2D eigenbrain (2D-EB). All the 3D brains were transferred to a feature space, which encoded the variation among known 3D brain images. The feature space was named as the 3D-EB, and defined as eigenvectors on the set of 3D brains. We compared four different classifiers: feed-forward neural network, support vector machine (SVM) with linear kernel, polynomial (Pol) kernel, and radial basis function kernel.

Results:

The 50x10-fold stratified cross validation experiments showed that the proposed 3D-EB is better than the 2D-EB. SVM with Pol kernel performed the best among all classifiers. Our “3D-EB + Pol-SVM” achieved an accuracy of 92.81% ± 1.99% , a sensitivity of 92.07% ± 2.48% , a specificity of 93.02% ± 2.22% , and a precision of 79.03% ± 2.37% . Based on the most important 3D-EB U₁, we detected 34 brain regions related with AD. The results corresponded to recent literature.

Conclusions:

We validated the effectiveness of the proposed 3D-EB by detecting subjects and brain regions related to AD.

Keywords

Alzheimer’s disease classification detection eigenbrain machine learning magnetic resonance imaging polynomial kernel prediction support vector machine

INTRODUCTION

Alzheimer’s disease (AD) is a chronic neurodegenerative disease that affects the aging progress [1]. It accounts for about 60% of dementia that causes severe problems in thinking, memory, and behavior [2, 3]. Symptoms usually lead to slow neural deterioration, and have negative influences on daily life until death [4 –6]. The report shows that there is currently no treatment for AD [7]. In 2006, 26.6 million people worldwide were reported to be suffering from AD. In 2050, AD is predicted to affect 1 in 85 people worldwide, and at least 43% of cases will need high level care [8]. As the population in the world ages, the people with AD bring heavy burdens and negative impacts to both their families and the society. In the United States, the healthcare cost for people with AD is about 100 billion dollars per year, and it is predicted to increase to a trillion dollars per year by 2050.

Early and accurate detection of AD is beneficial for the development and management of the disease [9, 10]. Currently, many neurologists and medical researchers have been contributing considerable time to researching methods to allow for early detection of AD, and promising results have been continually achieved [11]. Magnetic resonance imaging (MRI) can provide high quality images of the anatomical structures of the human body, especially of the brain [12, 13]. Using MRI, clinical diagnosis and biomedical research can obtain more information about AD [14, 15]. The diagnostic use of MRI has been greatly improved due to the automated and accurate labeling of MR images [16 –18], which plays a significant role in detecting AD in related subjects from healthy and elderly controls (HC) [19].

Earlier, most diagnosis work was carried out manually or semi-manually in order to measure a priori region of interest (ROI) of magnetic resonance (MR) images, based on the fact that patients with AD suffer more cerebral atrophy compared to HCs [20, 21]. Most of these ROI-based analyses focused on the shrinking of the cortex and hippocampus, and enlarged ventricles. However, ROI-based methods are not feasible in hospitals [22, 23].

Therefore, multivariate approaches were proposed as alternative methods to ROI-based methods. The advantage of using multivariate approaches is that they are data driven, which means that the analyses are fully based on the data without needing any prior knowledge. Meanwhile, the interactions among voxels and error effects can be assessed statistically [24]. However, multivariate approaches have been difficult in making statistical inferences about regionally specific changes due to the curse of the dimension, the small sample size, and the lack of capacity [25].

In this study, we proposed a novel method of the three-dimensional eigenbrain (3D-EB), which is an extension of the two-dimensional eigenbrain (2D-EB) and was inspired from the eigenface. Eigenface is an effective tool in handling the face recognition problem; Lopez et al. [26] and Alvarez et al. [27] developed eigenbrain on the basis of eigenface for single photon emission computed tomography (SPECT) images. Our past work was the first to apply eigenbrain to MR images. However, AD imaging data is volumetric, so it was necessary to extend 2D-EB to 3D-EB. This paper aimed to finish this task. Furthermore, we developed a novel system for the MR image classification based on 3D-EB, with the goal of developing a computer-aided diagnosis (CAD) system, which can be applied in the early detection of AD-related brain regions and subjects. Our purpose was to provide an assisting tool for clinicians.

Table 9 in the appendix lists all the acronyms and their meanings in this study for easier reading.

LITERATURE REVIEW

In the past, most diagnostic work was done by measuring the ROI of brain MR images [28, 29], due to the fact that AD brains have larger ventricles, smaller hippocampus, and smaller cortex than normal brains. However, the ROI-based methods suffered from some shortcomings: (i) ROI methods need a priori information and expert knowledge. (ii) The manual detection accuracy relied on the experience of physicians and interpreters [30]. (iii) The interaction among the voxels was difficult to implement [31]. (iv) It was necessary to explore other potential regions that may be connected to AD. (v) Automatic segmentation of ROI was not feasible in practice, and examiners tended to segment the brain manually [32].

Recently, a new type of method, the “whole-brain analysis”, is gaining popularity because it considers all the voxels in the brain as a whole. It is not necessary to segment the brain beforehand, and it does not need any biomarkers for the classification task. The main disadvantage is the curse of dimensionality, which can be solved with high-speed computers, which have become very inexpensive [25]. Whole-brain analysis heavily relies on pure computation, and it can only be accomplished by computer scientists after physicians help label the data as either AD or healthy. Generally, whole-brain analysis treats the whole brain as a ROI, and it consists of two phases: feature extraction and classification.

Currently, many methods were proposed in order to extract efficient features from brains with AD or other types of diseases. In addition, various classification models and methods exist. El-Dahshan et al. [33] extracted features of the approximation and detail coefficients of 3-level discrete wavelet transform (DWT). They used the classifiers based on the artificial neural network (ANN) and K-nearest neighbors. Plant et al. [34] used brain region cluster (BRC). They suggested using information gain (IG) to evaluate the significance of a voxel, and employed clustering algorithm to find groups of adjacent voxels with a high discriminatory power. They used support vector machine (SVM), voting feature intervals (VFI), and Bayes statistics for pattern classification. Park [35] employed manifold learning for classification. Their use of the first and second distance measure brought 18% and 46% error rates for classification between patients with AD and normal patients, respectively. Chaves et al. [36] utilized large margin-based methodology for AD detection in SPECT and positron emission tomography images. Zhang et al. [37] suggested that removing the spider-web-plot yielded the same classification performance. Savio and Grana [38] presented a novel method that employed the deformation-based morphometry method. They tested five features and found three features performed excellently as trace of Jacobian matrix (TJM), modulated grey matter (MGM), and geodesic anisotropy (GEODAN). Furthermore, they utilized Pearson’s correlation (PEC), Welch’s t-test (WTT), and Bhattacharyya distance (BD), to measure the significance of each voxel site. Kalbkhani et al. [39] employed the generalizing autoregressive conditional heteroscedasticity (GARCH) statistical model for the detail coefficients of 2-level DWT, and took the parameters of the GARCH model as the primary feature vector. They tested the kernel neural network and SVM models. Wang et al. [40] utilized the undersampling (US) technique on a 3D image. They used singular value decomposition (SVD) to select features. Then, they combined kernel SVM (KSVM) with decision tree. Harikumar and Kumar [41] analyzed the performance of ANN in terms of classification of medical images, using wavelets as feature extractor. They achieved a classification accuracy of 96% . Zhang et al. [18] employed discrete wavelet packet transform. They utilized Tsallis entropy to get wavelet packet entropy features. They introduced a generalized eigenvalue proximal SVM (GEPSVM). Nazir et al. [42] used filters for the removal of noises, and extracted color moments as mean features. They were able to achieve an overall accuracy of 91.8% . Zhang et al. [43] employed the eigenbrain to extract features, and then they employed the WTT in order to reduce features. They proposed using SVM with the radial basis function (RBF). Damodharan and Raghavan [44] combined tissue segmentation and neural network for brain tumor detection. Zhang et al. [45] proposed a novel classification system that implemented 3D-DWT to extract wavelet coefficients in the volumetric image. Zhang et al. [46] proposed combining stationary wavelet transform with GEPSVM. Farzan et al. [47] used longitudinal percentage of brain volume changes in a two-year follow up and its intermediate counterparts in early 6-month and late 18-month as features. Munteanu et al. [48] employed proton magnetic resonance spectroscopy data, with the goal of detecting mild cognitive impairment and AD. They used a single-layer perceptron with only two spectroscopic voxel volumes obtained in the left hippocampus, with an AUROC value of 0.866. Savio and Grana [49] utilized local activity features, such as regional homogeneity, to develop a CAD for schizophrenia with resting-state functional MRI. Zhang et al. [50] combined wavelet entropy with Hu moment invariants. The feature number in total was 14. They also used GEPSVM as the classifier. Zhang and Wang [51] proposed a 2D displacement field (DF) method to approximate the deformation of healthy control to AD. Gorji and Haddadnia [52] employed pseudo Zernike moment for the diagnosis of mild cognitive impairment from AD and HC. They employed two types of ANNs based on pattern recognition and learning vector quantization networks. Wang et al. [53] estimated 3D displacement field from structural MRI and used it asfeatures.

The method used in this paper is different from those described above. Our method was based on the eigenbrain, which was derived from eigenface that could handle the face recognition problem efficiently. We extended eigenbrain from 2D to 3D, so it can process volumetric data in a more efficient way.

MATERIALS AND METHODS

Materials

The dataset was downloaded from “Open Access Series of Imaging Studies (OASIS)” [54], which consisted of 416 subjects aged from eighteen to ninety-six. All of the subjects are right-handed. Because some of the samples contained partial data, we selected the samples that had complete data information. Finally, we selected 126 samples (28 ADs and 98 HCs) from the dataset. The demographic details are reported in Table 1. Following common convention, the Clinical Dementia Rating was interpreted as the target (label). Note that subjects either with missing records or less than sixty years old were removed.

Co-registration and brain-masking

In order to increase the signal-to-noise ratio, all 3D MR brain images were motion-corrected and spatially co-registered to the Talairach space in order to generate an averaged image, and then they were brain masked.

Figure 1 offers an example of preprocessing on 3D images with resolution of 1 mm × 1 mm × 1.25 mm. The motion-correction were registered in the 3D images of three scans, and then used to generate an image in the original acquisition space with resampling to 1 mm × 1 mm × 1 mm. Finally, the averaged image was normalized to the Talairach coordinate space, and then brain-extracted (Fig. 1).

Eigenface

The AD detection problem is similar to the face detection problem. The latter can be solved by the famous tool, “eigenface” [55]. The eigenface extracts features from a face image, encodes them, and then compares the coding of one face with a database of coding models.

Mathematically, eigenface uses the eigenvectors of the covariance matrix of the set of face images, which can be treated as features that characterize the changes between face images. Each image contributes to one eigenvector more or less. The eigenvector is displayed as a ghostly face named as eigenface. Figure 2 shows the eigenfaces generated from the “Yale Face Database [56]”.

Eigenbrain

Eigenbrain is inspired from the eigenface, and was used for pathological brain detection with the restraint that diseased brains should have obvious structural deformities distinctive from the healthy controls. AD subjects met this restraint, because they have severe atrophy in the cerebral cortex, severely enlarged ventricles, shrinkage of hippocampus, and many other regions reported in the literature.

In the past, Alvarez et al. [27] employed 2D-EB to detect AD. Lopez et al. [26] utilized 2D-EB and Bayesian classification rules. Zhang et al. [43] utilized 2D-EB and KSVM. All of the above methods achieved excellent results, but they all used 2D-EB. Figure 3 shows an illustration of 2D-EB.

The difference between past references and this work is that we extended 2D-EB to 3D-EB. The advantage of 3D-EB is that it can handle the 3D object more efficiently than the 2D-EB, but its disadvantage is that the costs for computing resources is too large.

Three-dimensional eigenbrain

The eigenbrains were limited to 2D, because their methods originated from eigenfaces that solved the face recognition problem, and face images are always in 2D. This caused a problem because the eigenbrain was unable to be directly applied to 3D brain imaging data. Therefore, the scholars extracted the most important slices and performed the eigenbrain operations over them. This technique did not consider the intervolved contents between slices, and therefore we proposed to extend eigenbrain to a novel 3D-EB. Figure 4 shows the development from eigenface to 3D-EB.

The implementation of 3D-EB followed traditional techniques of principal component analysis (PCA). X = [X₁, X₂, … , X_N]^T was assumed as a given data matrix of size N × A, where N stands for the number of samples and A for the number of attributes. In this task, the 3D brain image is a size of 176 × 208 × 176 = 6,443,008, so we needed to vectorize it to a row vector of 6,443,008 elements (A = 6,443,008).

First, we normalized the dataset matrix X to Z = [Z₁, Z₂, … , Z_N]^T which was mean-centered by subtracting its mean value and unit-variance scaled by dividing the difference by its standard deviation $Z \leftarrow \frac{X - μ (X)}{σ (X)}$ (1)

Next, we estimated the covariance matrix C with size of A × A by $C \leftarrow \frac{1}{N - 1} Z^{T} Z$ (2)

Here we used N–1 instead of N to produce an unbiased estimating value of the variance.

Third, we performed the eigendecomposition of C in the form of $C = U Λ U^{- 1}$ (3) in which, U is a matrix of size A × N, whose columns are the eigenvectors of the covariance matrix C . Matrix Λ is a diagonal matrix of size N × N, whose diagonal elements are eigenvalues of C . Each element corresponds to an eigenvector of A. We sort the eigenvalue matrix Λ and eigenvector matrix U in the order of decreasing value u₁> u₂> … > u_N.

In order to view the ith 3D-EB u(i), the ith column of U was re-constructed to a 3D brain. As the ith column of U has 6,443,008 elements, the re-constructed image was 176 × 208 × 176. $u (i) = reshape (U (:, i))$ (4)

In case N≪A (remember that N = 126 and A = 6,443,008 in this work), the computation cost of eigendecomposition of Equation (3) was enormous. However, it could be accelerated by replacing C in Equation (2) with C^′ $C^{'} \leftarrow \frac{1}{N - 1} {ZZ}^{T}$ (5) where, the size of C^′ is N × N, which can be used to significantly reduce the computation cost. The flowchart of calculating eigenbrain is shown in Figure 5.

The eigenvalue matrix Λ represents the distribution of energy in the source data among each of the eigenbrains. The eigenbrain matrixes U = [U₁, U₂, … , U_N]^T formed a basis for original data. From another point of view, the 3D-EB projected the brain with a size of A to a short eigenvalue vector with a size of N. This procedure can also be treated as an encoding implementation. Compared to 2D-EB, the 3D-EB can extract important features along the slice direction, and it does not need to perform eigenbrain at each slice.

Using 3D-EB to classify new brains

Once the 3D-EBs was created, identification became a pattern recognition task. In the training phase, we needed to transform each training brain X_n into its eigenbrain score vector S_n by $S_{n} = [S_{n} (1), S_{n} (2), \dots, S_{n} (k), \dots, S_{n} (N)]$ (6) where S_n(k) represents the score of the n-th brain projected to k-th 3D-EB. It can be calculated by $S_{n} (k) = u_{k}^{T} (\frac{X_{n} - μ (X)}{σ (X)}) = u_{k}^{T} Z_{n}$ (7)

The classifier was generated on the score set of training brains. Figure 6 shows how the 3D-EB helped detect new brains. In the training phase, we first extracted mean (μ) and variance (σ) from the brain images, and employed μ and σ to normalize those brains. Then, we extracted eigenbrains U from the normalized brains, and employ U to obtain the scores of each brain. Finally, based on the scores we constructed the classifiers. In the prediction phase, we utilized the μ & σ obtained in the training phase to normalize the new brain, and the obtained eigenbrain U to extract the score, and then we established classifier to produce the predicted result.

Two canonical supervised classifiers were used: feed-forward neural network (FNN) [57] and SVM. Both are successfully applied in various academic and industrial applications. The former is more suitable for small-size problems, and the latter works well in classifying nonlinear data.

To further improve the classification performance, two kernel approaches were embedded to the SVM. They were polynomial (Pol) kernel and RBF kernel. The hybrid SVMs are Pol-SVM and RBF-SVM. The kernels mapped original data into a higher-dimension feature space [58], and the transformation may be used to provide non-linear segmentation in original space.

In total, four CAD systems were tested in this study, i.e., 3D-EB + FNN, 3D-EB + SVM, 3D-EB + Pol-SVM, and 3D-EB + RBF-SVM. Their performance was tested by professional statistical analysis.

Statistical analysis

In this paper, we employed the 10-fold stratified cross validation (SCV) considering the trade off between computational cost and reliable estimates. The 10-fold SCV was repeated 50 times, i.e., we carried out a 50 × 10-fold SCV. For each run, we took the accuracy, sensitivity, specificity, and precision as the measurements of the performance (Table 2). A correctly identified AD case was taken as a true positive, following the common convention. Based on 50 repetitions, the final measures of both the mean and standard deviation (SD) of the four measures were reported. Those measures were used for AD detection, other than the AD-related brain regiondetection.

Region detection

Remember the values in Λ are usually sorted in decreasing order, so U₁ is the most important 3D-EB and was used for region detection. The voxels in U₁ were used to select distinguishing voxels. The four-step region-detection procedure is listed in Table 3.

RESULTS

The programs were in-house developed by Matlab 2014a, and run on IBM laptop with 3 GHz Intel i3 dual-processor with 8GB RAM. The results can be repeated on any machine where Matlab is available.

3D-Eigenbrain

We obtained the 3D-EBs, among which the most important one U₁ was shown in Fig. 7 with a slice from 20 to 185 with an increment of 5. The eigenbrain is not a real brain but can be regarded as a “ghost brain” that reveals the distinguishing voxels between AD and HC.

Comparison between FNN and SVM

In the second experiment, we compared the FNN with SVM based on 50 × 10-fold SCV. Features were selected as the scores by 3D-EB, namely, we compared 3D-EB + FNN with 3D-EB + SVM. The results are listed in Table 4.

3D-EB versus 2D-EB

In the third experiment, we compared the proposed 3D-EB with traditional 2D-EB. The results of 2D-EB were obtained from reference [43]. Note that 2D-EB is followed by a WTT since 2D-EB obtains too many features. Table 5 compared the features of “2D-EB + WTT” with 3D-EB. Both classifiers were chosen as SVM.

SVM versus KSVMs

In the fourth experiment, we demonstrated the effect of kernel methods. We compared the standard SVM with SVMs with Pol kernel and RBF kernel with features of scores obtained by 3D-EB, viz., we will compare the three proposed CAD systems: 3D-EB + SVM, 3D-EB + RBF-SVM, and 3D-EB + Pol-SVM. The comparison results are shown in Table 6.

Comparison to state-of-the-art

In the fifth experiment, we compared the best proposed method “3D-EB + Pol-SVM” with 11state-of-the-art approaches. The classification comparison was listed in Table 7.

Region detection

In the final experiment, we carried out the region detection procedure described in Table 3. Figure 8 shows the result. The “autumn” color map was added to the distinguishing points. Red represents a small value while yellow represents a large value.

Region labeling

Here we reported the regions interpreted by 3D-EB in Table 8, where BA represented Brodmann area.

DISCUSSION

After comparing the results of FNN with SVM in Table 4, we found the FNN achieved an accuracy of 90.98% ± 4.02% , sensitivity of 89.93% ± 3.68% , specificity of 91.29% ± 3.21% , and precision of 74.67% ± 3.35% . The SVM achieved an accuracy of 92.37% ± 2.10% , sensitivity of 91.57% ± 2.15% , specificity of 92.59% ± 2.33% , and precision of 77.93% ± 1.89% . It is obvious that SVM obtained better results than FNN did. Our results fell in line with recent studies showing that SVM performs better than FNN for small-size problem [85, 86]. In this study, we only included 126 subjects; hence, SVM will give better results than FNN.

Results in Table 5 show that 3D-EB + SVM obtained higher values in terms of accuracy, sensitivity, specificity, and precision than “2D-EB + WTT”. This validated the effectiveness of 3D-EB, and supported our extension from 2D to 3D for the eigenbrain.

From Table 6 we saw that SVM achieved an accuracy of 92.37% ± 2.10% , sensitivity of 91.57% ± 2.15% , specificity of 92.59% ± 2.33% , and precision of 77.93% ± 1.89% . RBF-SVM obtained results worse than SVM (accuracy of 88.54% ± 2.37% , sensitivity of 88.36% ± 3.07% , specificity of 88.59% ± 3.10% , and precision of 68.88% ± 2.52% ). Finally, the Pol-SVM obtained the best result of accuracy of 92.81% ± 1.99% , sensitivity of 92.07% ± 2.48% , specificity of 93.02% ± 2.22% , and precision of 79.03% ± 2.37% . Those results showed that the Pol-SVM performs better than linear SVM and RBF-SVM. It gave us a suggestion that RBF was not the best for all applications. In some conditions, the Pol kernel may provide better performance than RBF kernel.

Classification results in Table 7 compared 3D-EB + Pol-SVM with state-of-the-art approaches, which used different statistical settings, making comparison difficult. The data showed that our results are better than or comparable to other approaches. From the point of average accuracy, BRC + IG + SVM also achieved 90% [34], BRC + IG + Bayes achieved 92% [34], BRC + IG + VFI achieved 78.00% [34], MGM + PEC + SVM achieved 92.07% [38], GEODAN + BD + SVM of 92.09% [38], TJM + WTT + SVM of 92.83% [38], US + SVD-PCA + SVM-DT achieved 90% [40], EB + WTT + SVM [43] achieved 91.47% , EB + WTT + SVM + RBF [43] achieved 86.71% , EB + WTT + SVM + Pol [43] achieved 92.36% , and DF + PCA + SVM [51] achieved 88.27% . Finally, the proposed 3D-EB + Pol-SVM achieved an average accuracy of 92.81% .

Results in Table 8 show that U₁ interpreted the discriminant voxels between AD and HC involving the following 34 regions: Anterior Cingulate (BA-9, BA-10, BA-24, BA-25, BA-32, BA-33, Corpus Callosum) [59], Caudate (head, body, and tail) [60], Cingulate Gyrus (BA-6, BA-9, BA-23, BA-24, BA-31, BA-32) [61], Claustrum [62], Culmen [63], Cuneus (BA-18, BA-30) [64], Fusiform Gyrus [65], Inferior Frontal Gyrus (BA-A, BA-10, BA-11, BA-25, BA-44, BA-45, BA-46, 47) [66], Inferior Parietal Lobule (BA-40) [67], Inferior Temporal Gyrus (BA-20, BA-21, BA-37) [68], Insula (BA-13, BA-22, BA-40, BA-41, BA-47) [69], Lateral Ventricle [70], Lentiform Nucleus (Putamen, Lateral Globus Pallidus, Medial Globus Pallidus) [60], Lingual Gyrus (BA-18, BA-19) [71], Medial Frontal Gyrus (BA-6, BA-8, BA-9, BA-10, BA-11, BA-25, BA-32) [72], Middle Frontal Gyrus (BA-6, BA-8, BA-9, BA-10, BA-11, BA-46, BA-47) [59], Middle Occipital Gyrus [71], Middle Temporal Gyrus (BA-21, BA-38) [73], Paracentral Lobule (BA-6, BA-31) [72], Parahippocampal Gyrus (Amygdala, BA-19, BA-20, BA-28, BA-30, BA-34 BA-35, BA-36, BA-37, BA-38, Hippocampus) [74], Postcentral Gyrus (BA-2, BA-40, BA-43) [72], Posterior Cingulate (BA-18, BA-23, BA-29, BA-30, BA-31) [75], Precentral Gyrus (BA-4, BA-6, BA-9, BA-44) [72], Precuneus (BA-7, BA-31) [72], Rectal Gyrus (BA-11) [76], Subcallosal Gyrus (BA-11, BA-25, BA-34, BA-47) [77], Sub-Gyral (BA-4, BA-6, BA-8, BA-10, BA-20, BA-21, BA-32, BA-46, BA-47, Corpus Callosum, Hippocampus) [78], Superior Frontal Gyrus (BA-6, BA-8, BA-9, BA-10, BA-11) [79], Superior Temporal Gyrus (BA-13, BA-22, BA-38, BA-41, BA-42) [77], Supramarginal Gyrus [80], Thalamus (Anterior Nucleus, Lateral Dorsal Nucleus, Lateral Posterior Nucleus, Mammilary body, Medial Dorsal Nucleus, Midline Nucleus, Pulvinar, Ventral Anterior Nucleus, Ventral Lateral Nucleus, Ventral Posterior Lateral Nucleus, Ventral Posterior Medial Nucleus) [69], Third Ventricle [81], Transverse Temporal Gyrus (BA-41) [82], Tuber [83], and Uncus (Amygdala, BA-20, BA-28, BA-34, BA-36, BA-38) [84].

Above regions were detected by 3D-EB since those regions have obvious differences in the U₁ between HC and AD. However, those regions may or may not be the source of AD. From another point of view, Table 8 demonstrates the power of 3D-EB. Our approach employed U₁ to find distinguishing regions, and the findings covered 34 regions reported by a mass of recent literatures, based on different feature extraction methods and advanced imaging technologies.

CONCLUSION AND FUTURE RESEARCH

The contributions of this study are as follows: (i) We extended the 2D-EB to 3D-EB, and proved its effectiveness; (ii) We proposed four novel CAD systems based on different hybrid 3D-EB that can detect AD related brain regions and subjects, and validated the best was 3D-EB + Pol-SVM that had a classification accuracy of 92.81 ± 1.99, which was comparable to state-of-the-art methods. (iii) We proved 34 brain regions were related to AD.

In the future, we will carry out the following approaches: (i) Other kernels for SVM will be tested, and we will try to find alternative pattern recognition tools for KSVM, such as fuzzy SVM [87]. (ii) The proposed 3D-EB can be combined with other features, and an improvement of classification accuracy is expected. (iii) Swarm intelligence methods, such as particle swarm optimization [88], artificial bee colony [89], ant colony optimization [90], biogeography-based optimization [91], and genetic pattern search [92], may be applied to find the optimal kernel parameters. (iv) The proposed method will be tested on images generated from other modalities: computed tomography [93], ultrasound, spectrum imaging [94], etc. (v) Classification may be carried out on the sparsity domain [95, 96].

Footnotes

ACKNOWLEDGMENTS

This paper was supported by NSFC (61273243, 51407095), Natural Science Foundation of Jiangsu Province (BK20150982, BK20150983), Priority Academic Program Development of Jiangsu Higher Education Institutions (PAPD), Jiangsu Key Laboratory of 3D Printing Equipment and Manufacturing (BM2013006), Key Supporting Science and Technology Program (Industry) of Jiangsu Province (BE2012201, BE2013012-2, BE2014009-3), Program of Natural Science Research of Jiangsu Higher Education Institutions (13KJB460011, 14KJB520021), Special Funds for Scientific and Technological Achievement Transformation Project in Jiangsu Province (BA2013058), Nanjing Normal University Research Foundation for Talented Scholars (2013119XGQ0061, 2014119XGQ0080), Education Reform Project in NJNU (18122000090615), Open Fund of Guangxi Key Laboratory of Manufacturing System & Advanced Manufacturing Technology (15-140-30-008K).

The authors acknowledge their gratitude to the OASIS dataset that came from NIH grants P50AG05681, P01 AG03991, R01 AG021910, P50 MH071616, U24 RR021382, and R01 MH56584.

Authors’ disclosures available online ().

References

Cao

, Zhao

, Wei

, Chen

, Yang

, Ou

, Wu

, Shang

(2015) The global cognition, frontal lobe dysfunction and behavior changes in Chinese patients with multiple system atrophy. PLoS One 10, 14.

Guerra

, Hua

, Wunsch

(2015) Risk of a diagnosis of dementia for elderly Medicare beneficiaries after intensive care. Anesthesiology 123, 1105–1112.

Weakley

, Williams

, Schmitter-Edgecombe

, Cook

(2015) Neuropsychological test selection for cognitive impairment classification: A machine learning approach. J Clin Exp Neuropsychol 37, 899–916.

Borroni

, Cosseddu

, Pilotto

, Premi

, Archetti

, Gasparotti

, Cappa

, Padovani

(2015) Early stage of behavioral variant frontotemporal dementia: clinical and neuroimaging correlates. Neurobiol Aging 36, 3108–3115.

Talib

, Hototian

, Joaquim

HPG

, Forlenza

, Gattaz

(2015) Increased iPLA2 activity and levels of phosphorylated GSK3B in platelets are associated with donepezil treatment in Alzheimer’s disease patients. Eur Arch Psychiatry Clin Neurosci 265, 701–706.

Tabaton

, Zhu

, Perry

, Smith

, Giliberto

(2010) Signaling effect of amyloid-beta(42) on the processing AbetaPP. Exp Neurol 221, 18–25.

Belleville

, Fouquet

, Duchesne

, Collins

, Hudon

, Grp

C-Q

(2014) Detecting early preclinical Alzheimer’s disease via cognition, neuropsychiatry, and neuroimaging: qualitative review and recommendations for testing. J Alzheimers Dis 42, S375–S382.

Tabaton

, Gambetti

(2006) Soluble amyloid-beta in the brain: The scarlet pimpernel. J Alzheimers Dis 9, 127–132.

Tosto

, Gasparini

, Brickman

, Letteri

, Renie

, Piscopo

, Talarico

, Canevelli

, Confaloni

, Bruno

(2015) Neuropsychological predictors of rapidly progressive Alzheimer’s disease. Acta Neurol Scand 132, 417–422.

10.

Wang

, Dong

, Du

, Ji

, Yan

, Yang

, Wang

, Feng

, Phillips

(2015) Feed-forward neural network optimized by hybridization of PSO and ABC for abnormal brain detection. Int J Imaging Syst Technol 25, 153–164.

11.

Tabaton

, Odetti

, Cammarata

, Borghi

, Monacelli

, Caltagirone

, Bossu

, Buscema

, Grossi

(2010) Artificial neural networks identify the predictive values of risk factors on the conversion of amnestic mild cognitive impairment. J Alzheimers Dis 19, 1035–1040.

12.

Dong

, Phillips

, Wang

, Ji

, Yang

(2015) Exponential wavelet iterative shrinkage thresholding algorithm for compressed sensing magnetic resonance imaging. Inf Sci (Ny) 322, 115–132.

13.

Zhang

, Dong

, Ji

(2014) An improved reconstruction method for CS-MRI based on exponential wavelet transform and iterative shrinkage/thresholding algorithm. J Electromagn Waves Appl 28, 2327–2338.

14.

Salvatore

, Cerasa

, Battista

, Gilardi

, Quattrone

, Castiglioni

, Alzheimer’s Disease Neuroimaging Initiative (2015) Magnetic resonance imaging biomarkers for the early diagnosis of Alzheimer’s disease: a machine learning approach. Front Neurosci 9, 13.

15.

, Dai

, Fan

, Ma

, Tang

, Zhang

(2015) Nanotheranostics: Congo red/Rutin-MNPs with enhanced magnetic resonance imaging and H2O2-responsive therapy of Alzheimer’s disease in APPswe/PS1de9 transgenic mice. Adv Mater 27, 5499–5505.

16.

, Shui

, Gen

, Zheng

(2015) Exponential wavelet iterative shrinkage thresholding algorithm with random shift for compressed sensing magnetic resonance imaging. IEEJ Trans Electr Electron Eng 10, 116–117.

17.

Goh

, Dong

, Zhang

, DiMauro

, Peterson

(2014) Mitochondrial dysfunction as a neurobiological subtype of autism spectrum disorder: Evidence from brain imaging. JAMA Psychiatry 71, 665–671.

18.

Dong

, Wang

, Ji

, Yang

(2015) Preclinical diagnosis of magnetic resonance (MR) brain images via discrete wavelet packet transform with tsallis entropy and generalized eigenvalue proximal support vector machine (GEPSVM). Entropy 17, 1795–1813.

19.

Jeurissen

, Leemans

, Sijbers

(2014) Automated correction of improperly rotated diffusion gradient orientations in diffusion weighted MRI. Med Image Anal 18, 953–962.

20.

Hamelin

, Bertoux

, Bottlaender

, Corne

, Lagarde

, Hahn

, Mangin

, Dubois

, Chupin

, de Souza

, Colliot

, Sarazin

(2015) Sulcal morphology as a new imaging marker for the diagnosis of early onset Alzheimer’s disease. Neurobiol Aging 36, 2932–2939.

21.

Caballero

MAA

, Brendel

, Delker

, Ren

, Rominger

, Bartenstein

, Dichgans

, Weiner

, Ewers

, Alzheimer’s

Disease Neuroimaging Initiative

(2015) Mapping 3-year changes in gray matter and metabolism in Abeta-positive nondemented subjects. Neurobiol Aging 36, 2913–2924.

22.

Toledo

, Bjerke

, Chen

, Rozycki

, Jack

, Weiner

, Arnold

, Reiman

, Davatzikos

, Shaw

, Trojanowski

(2015) Memory, executive, and multidomain subtle cognitive impairment Clinical and biomarker findings. Neurology 85, 144–153.

23.

Song

, Long

, Zuo

, Yu

, Liu

, Wang

, Han

, Jia

(2015) APOE effects on default mode network in chinese cognitive normal elderly: relationship with clinical cognitive performance. PLoS One 10, 11.

24.

Zhang

, Dong

, Phillips

, Ji

, Yang

(2015) Pathological brain detection in magnetic resonance imaging scanning by wavelet entropy and hybridization of biogeography-based optimization and particle swarm optimization. Prog Electromagn Res 152, 41–58.

25.

Álvarez

, Górriz

, Ramírez

, Salas-Gonzalez

, López

, Segovia

, Puntonet

, Prieto

(2009) Alzheimer’s diagnosis using eigenbrains and support vector machines. In Bio-Inspired Systems: Computational and Ambient Intelligence, Cabestany

, Sandoval

, Prieto

, Corchado

, eds. Springer Berlin Heidelberg, pp. 973–980.

26.

Lopez

, Ramirez

, Gorriz

, Alvarez

, Salas-Gonzalez

, Segovia

, Puntonet

(2009) Automatic system for Alzheimer’s disease diagnosis using eigenbrains and Bayesian classification rules. In Bio-Inspired Systems: Computational and Ambient Intelligence , Pt 1, Cabestany

, Prieto

, Sandoval

, Corchado

, eds. Springer-Verlag, Berlin, pp. 949–956.

27.

Alvarez

, Gorriz

, Ramirez

, Salas-Gonzalez

, Lopez

, Puntonet

, Segovia

(2009) Alzheimer’s diagnosis using eigenbrains and support vector machines. Electron Lett 45, 342–343.

28.

Kubota

, Ushijima

, Nishimura

(2006) A region-of-interest (ROI) template for three-dimensional stereotactic surface projection (3D-SSP) images: Initial application to analysis of Alzheimer disease and mild cognitive impairment. Int Congr Ser 1290, 128–134.

29.

Anagnostopoulos

, Giannoukos

, Spenger

, Simmons

, Mecocci

, Soininen

, Kloszewska

, Vellas

, Lovestone

, Tsolaki

(2013) Classification models for Alzheimer’s disease detection. In Iliadis L, Papadopoulos H, Jayne C, eds., pp. Engineering Applications of Neural Networks, Pt Ii 193–202.

30.

Yang

, Zhang

, Yang

, Ji

, Dong

, Wang

, Feng

, Wang

(2015) Automated classification of brain images using wavelet-energy and biogeography-based optimization. Multimed Tools Appl, doi: 10.1007/s11042-015-2649-7

31.

Lee

, Park

, Han

(2013) Classification of diffusion tensor images for the early detection of Alzheimer’s disease. Comput Biol Med 43, 1313–1320.

32.

Zhang

, Wang

, Ji

, Dong

(2013) An MR brain images classifier system via particle swarm optimization and kernel support vector machine. ScientificWorldJournal 2013, 130134.

33.

El-Dahshan

ESA

, Hosny

, Salem

ABM

(2010) Hybrid intelligent techniques for MRI brain images classification. Digit Signal Process 20, 433–441.

34.

Plant

, Teipel

, Oswald

, Böhm

, Meindl

, Mourao-Miranda

, Bokde

, Hampel

, Ewers

(2010) Automated detection of brain atrophy patterns based on MRI for the prediction of Alzheimer’s disease. Neuroimage 50, 162–174.

35.

Park

(2012) Comparison of distance measures for manifold learning: Application to Alzheimer’s brain scans. J Korean Phys Soc 61, 1148–1155.

36.

Chaves

, Ramirez

, Gorriz

, Illan

, Gomez-Rio

, Carnero

, Alzheimer’s

Disease Neuroimaging Initiative

(2012) Effective diagnosis of Alzheimer’s disease by means of large margin-based methodology. BMC Med Inform Decis Mak 12, 17.

37.

Zhang

, Ji

, Wang

(2015) Effect of spider-web-plot in MR brain image classification. Pattern Recognit Lett 62, 14–16.

38.

Savio

, Grana

(2013) Deformation based feature selection for computer aided diagnosis of Alzheimer’s disease. Expert Syst Appl 40, 1619–1628.

39.

Kalbkhani

, Shayesteh

, Zali-Vargahan

(2013) Robust algorithm for brain magnetic resonance image (MRI) classification based on GARCH variances series. Biomed Signal Process Control 8, 909–919.

40.

Wang

, Dong

, Ji

, Zhang

(2014) Classification of Alzheimer disease based on structural magnetic resonance imaging by kernel support vector machine decision tree. Prog Electromagn Res 144, 171–184.

41.

Harikumar

, Kumar

(2015) Performance analysis of neural networks for classification of medical images with wavelets as a feature extractor. Int J Imaging Syst Technol 25, 33–40.

42.

Nazir

, Wahid

, Khan

(2015) A simple and intelligent approach for brain MRI classification. J Intell Fuzzy Syst 28, 1127–1135.

43.

Zhang

, Dong

, Phillips

, Wang

, Ji

, Yang

, Yuan

T-F

(2015) Detection of subjects and brain regions related to Alzheimer’s disease using 3D MRI scans based on eigenbrain and machine learning. Front Comput Neurosci 9, 66.

44.

Damodharan

, Raghavan

(2015) Combining tissue segmentation and neural network for brain tumor detection. Int Arab J Inf Technol 12, 42–52.

45.

Wang

, Phillips

, Dong

, Ji

, Yang

(2015) Detection of Alzheimer’s disease and mild cognitive impairment based on structural volumetric MR images using 3D-DWT and WTA-KSVM trained by PSOTVAC. Biomed Signal Process Control 21, 58–73.

46.

Zhang

, Dong

, Liu

, Wang

, Ji

, Zhang

, Yang

(2015) Magnetic resonance brain image classification via stationary wavelet transform and generalized eigenvalue proximal support vector machine. J Med Imaging Health Inform 5, 1395–1403.

47.

Farzan

, Mashohor

, Ramli

, Mahmud

(2015) Boosting diagnosis accuracy of Alzheimer’s disease using high dimensional recognition of longitudinal brain atrophy patterns. Behav Brain Res 290, 124–130.

48.

Munteanu

, Fernandez-Lozano

, Abad

, Fernandez

, Alvarez-Linera

, Hernandez-Tamames

, Pazos

(2015) Classification of mild cognitive impairment and Alzheimer’s disease with machine-learning techniques using H-1 Magnetic Resonance Spectroscopy data. Expert Syst Appl 42, 6205–6214.

49.

Savio

, Grana

(2015) Local activity features for computer aided diagnosis of schizophrenia on resting-state fMRI. Neurocomputing 164, 154–161.

50.

Wang

, Sun

, Phillips

(2015) Pathological brain detection based on wavelet entropy and Hu moment invariants. Biomed Mater Eng 26, 1283–1290.

51.

Wang

(2015) Detection of Alzheimer’s disease by displacement field and machine learning. PeerJ 3, e1251.

52.

Gorji

, Haddadnia

(2015) A novel method for early diagnosis of Alzheimer’s disease based on pseudo Zernike moment from structural MRI. Neuroscience 305, 361–371.

53.

Wang

, Zhang

, Liu

, Phillips

, Yuan

T-F

(2016) Detection of Alzheimer’s disease by three-dimensional displacement field estimation in structural magnetic resonance imaging. J Alzheimers Dis 50, 1–16.

54.

Ardekani

, Figarsky

, Sidtis

(2013) Sexual dimorphism in the human corpus callosum: An MRI study using the OASIS brain database. Cereb Cortex 23, 2514–2520.

55.

Chandrasiri

, Suzuki

, Watanabe

, Yamada

(2013) The effect of distinctiveness in recognizing average face: Human recognition and eigenface based machine recognition. IEICE Trans Inf Syst E96D, 514–522.

56.

Belhumeur

, Hespanha

, Kriegman

(1997) Eigenfaces vs. Fisherfaces: Recognition using class specific linear projection. IEEE Trans Pattern Anal Mach Intell 19, 711–720.

57.

Zhang

, Wang

S-H

, Ji

, Phillips

(2014) Fruit classification using computer vision and feedforward neural network. J Food Eng 143, 167–177.

58.

Fang

L-T

, Wu

L-N

, Zhang

Y-D

(2015) A novel demodulation system based on continuous wavelet transform. Math Probl Eng 2015, 1–9.

59.

Schultz

, Larson

, Oh

, Koscik

, Dowling

, Gallagher

, Carlsson

, Rowley

, Bendlin

, Asthana

, Hermann

, Johnson

, Sager

, LaRue

, Okonkwo

(2014) Participation in cognitively-stimulating activities is associated with brain structure and cognitive function in preclinical Alzheimer’s disease. Brain Imaging Behav 9, 729–736.

60.

Moller

, Dieleman

, van der Flier

, Versteeg

, Pijnenburg

, Scheltens

, Barkhof

, Vrenken

(2014) More atrophy of deep gray matter structures in frontotemporal dementia compared to Alzheimer’s disease. J Alzheimers Dis 44, 635–647.

61.

, Peng

, Mishra

, Ouyang

, Li

, Zhang

, Chen

, Liu

, Huang

(2014) Microstructure, length, and connection of limbic tracts in normal human brain development. Front Aging Neurosci 6, 228–.

62.

De Reuck

, Deramecourt

, Auger

, Durieux

, Cordonnier

, Devos

, Defebvre

, Moreau

, Caparros-Lefebvre

, Leys

, Maurage

, Pasquier

, Bordet

(2014) Iron deposits in post-mortem brains of patients with neurodegenerative and cerebrovascular diseases: a semi-quantitative 7.0 T magnetic resonance imaging study. Eur J Neurol 21, 1026–1031 .

63.

Shimoda

, Kimura

, Yokota

, Okubo

(2015) Comparison of regional gray matter volume abnormalities in Alzheimer’s disease and late life depression with hippocampal atrophy using VSRAD analysis: A voxel-based morphometry study. Psychiatry Res 232, 71–75.

64.

Graff-Radford

, Murray

, Lowe

, Boeve

, Ferman

, Przybelski

, Lesnick

, Senjem

, Gunter

, Smith

, Knopman

, Jack

, Dickson

, Petersen

, Kantarci

(2014) Dementia with Lewy bodies basis of cingulate island sign. Neurology 83, 801–809.

65.

Watson

, Colloby

, Blamire

, O’Brien

(2015) Assessment of regional gray matter loss in dementia with lewy bodies: a surface-based MRI analysis. Am J Geriatr Psychiatry 23, 38–46.

66.

Eliasova

, Anderkova

, Marecek

, Rektorova

(2014) Non-invasive brain stimulation of the right inferior frontal gyrus may improve attention in early Alzheimer’s disease: A pilot study. J Neurol Sci 346, 318–322.

67.

Wang

, Xia

, Dai

, Liang

, Song

, He

, Li

(2015) Differentially disrupted functional connectivity of the subregions of the inferior parietal lobule in Alzheimer’s disease. Brain Struct Funct 220, 745–762.

68.

Dowling

, Johnson

, Gleason

, Jagust

, Alzheimer’s

Disease Neuroimaging Initiative

(2015) The mediational effects of FDG hypometabolism on the association between cerebrospinal fluid biomarkers and neurocognitive function. Neuroimage 105, 357–368.

69.

, Liu

, Radua

, Li

, Han

, Sun

(2015) Meta-analytic Comparison Between PIB-PET and FDG-PET Results in Alzheimer’s Disease and MCI. Cell Biochem Biophys 71, 17–26 .

70.

Voevodskaya

, Simmons

, Nordenskjold

, Kullberg

, Ahlstrom

, Lind

, Wahlund

, Larsson

, Westman

(2014) The effects of intracranial volume adjustment approaches on multiple regional MRI volumes in healthy aging and Alzheimer’s disease. Front Aging Neurosci 6, 264.

71.

Lehmann

, Ghosh

, Madison

, Laforce

, Corbetta-Rastelli

, Weiner

, Greicius

, Seeley

, Gorno-Tempini

, Rosen

, Miller

, Jagust

, Rabinovici

(2013) Diverging patterns of amyloid deposition and hypometabolism in clinical variants of probable Alzheimer’s disease. Brain 136, 844–858.

72.

Kang

, Yoon

, Lee

(2013) Idiopathic normal-pressure hydrocephalus, cortical thinning, and the cerebrospinal fluid tap test. J Neurol Sci 334, 55–62.

73.

Aubry

, Shin

, Crary

, Lefort

, Qureshi

, Lefebvre

, Califano

, Shelanski

(2015) Assembly and interrogation of Alzheimer’s disease genetic networks reveal novel regulators of progression. PLoS One 10, 25.

74.

Eskildsen

, Coupé

, Fonov

, Pruessner

, Collins

(2015) Structural imaging biomarkers of Alzheimer’s disease: predicting disease progression. Neurobiol Aging 36, Suppl 1, S23–S31.

75.

Shinohara

, Fujioka

, Murray

, Wojtas

, Baker

, Rovelet-Lecrux

, Rademakers

, Das

, Parisi

, Graff-Radford

, Petersen

, Dickson

, Bu

(2014) Regional distribution of synaptic markers and APP correlate with distinct clinicopathological features in sporadic and familial Alzheimer’s disease. Brain 137, 1533–1549.

76.

Huang

, Yen

, Wu

, Chen

, Hung

, Kao

, Chen

, Ho

(2015) Brain metabolism of less-educated patients with Alzheimer dementia studied by positron emission tomography. Medicine (Baltimore) 94, e1252.

77.

Paakki

, Rahko

, Long

, Moilanen

, Tervonen

, Nikkinen

, Starck

, Remes

, Hurtig

, Haapsamo

, Jussila

, Kuusikko-Gauffin

, Mattila

, Zang

, Kiviniemi

(2010) Alterations in regional homogeneity of resting-state brain activity in autism spectrum disorders. Brain Res 1321, 169–179.

78.

Streitburger

, Moller

, Tittgemeyer

, Hund-Georgiadis

, Schroeter

, Mueller

(2012) Investigating structural brain changes of dehydration using voxel-based morphometry. PloS One 7, e44195.

79.

Chen

, Liu

, Zhang

, Xu

, Zhang

, Wei

, Zhang

(2014) Altered brain activation patterns under different working memory loads in patients with type 2 diabetes. Diabetes Care 37, 3157–3163.

80.

Quiroz

, Stern

, Reiman

, Brickhouse

, Ruiz

, Sperling

, Lopera

, Dickerson

(2013) Cortical atrophy in presymptomatic Alzheimer’s disease presenilin 1 mutation carriers. J Neurol Neurosurg Psychiatry 84, 556–561.

81.

Shingo

, Kanabayashi

, Kito

, Murase

(2013) Intracerebroventricular administration of an insulin analogue recovers STZ-induced cognitive decline in rats. Behav Brain Res 241, 105–111.

82.

Kim

, Lee

, Park

, Kim

, Bae

, Kim

, Im

(2012) Clinical implications of quantitative electroencephalography and current source density in patients with Alzheimer’s disease. Brain Topogr 25, 461–474.

83.

Schultz

, Ghebremedhin

, Braak

(1999) Sex-dependent cytoskeletal changes of the human hypothalamus develop independently of Alzheimer’s disease. Exp Neurol 160, 186–193.

84.

Bangen

, Nation

, Clark

, Harmell

, Wierenga

, Dev

, Delano-Wood

, Zlatar

, Salmon

, Liu

, Bondi

(2014) Interactive effects of vascular risk burden and advanced age on cerebral blood flow. Front Aging Neurosci 6, 159.

85.

Jiang

, Zan

, Zhu

, Liu

, Shao

(2013) Rectal sensation function rebuilding based on optimal wavelet packet and support vector machine. IET Sci Meas Technol 7, 139–144.

86.

Vong

, Wong

, Tam

, Zhang

(2011) Ignition pattern analysis for automotive engine trouble diagnosis using wavelet packet transform and support vector machines. Chin J Mech Eng 24, 870–878.

87.

Yang

, Zhang

, Phillips

, Yang

, Yuan

T-F

(2015) Identification of green, oolong and black teas in china via wavelet packet entropy and fuzzy support vector machine. Entropy 17, 6663–6682.

88.

Wang

S-H

, Phillips

, Ji

(2014) Binary PSO with mutation operator for feature selection using decision tree applied to spam detection. Knowl Based Syst 64, 22–31.

89.

L-N

, Wang

S-H

(2011) Magnetic resonance brain image classification by an improved artificial bee colony algorithm. Prog Electromagn Res 116, 65–79.

90.

Pang

, Jiang

, Wang

, Hu

, Liu

, Deng

, Huang

(2012) Gene order computation using Alzheimer’s DNA microarray gene expression data and the ant colony optimisation algorithm. Int J Data Min Bioinform 6, 617–632.

91.

Wang

, Ji

G-L

, Yang

, Wu

J-G

, Wei

(2015) Fruit classification by wavelet-entropy and feedforward neural network trained by fitness-scaled chaotic abc and biogeography-based optimization. Entropy 17, 5711–5728.

92.

Zhang

Y-D

, Ji

G-L

, Dong

Z-C

(2013) Genetic pattern search and its application to brain image classification. Math Probl Eng 2013, 580876.

93.

Chen

, Zhang

, Yang

, Cao

, Yang

, Chen

, Shu

, Luo

, Coatrieux

J-L

, Feng

(2015) Curve-like structure extraction using minimal path propagation with back-tracing. IEEE Trans Image Process 99, 1–16.

94.

Dong

, Zhang

, Liu

, Duan

, Kangarlu

, Peterson

(2014) Improving the spectral resolution and spectral fitting of 1H MRSI data from human calf muscle by the SPREAD technique. NMR Biomed 27, 1325–1332.

95.

, Peterson

, Dong

(2011) A two-level iterative reconstruction method for compressed sensing MRI. J Electromagn Waves Appl 25, 1081–1091.

96.

Zhang

, Peterson

, Ji

, Dong

(2014) Energy preserved sampling for compressed sensing MRI. Comput Math Methods Med 2014, 546814.