fbpx

Computer aided diagnosis of Alzheimer’s disease: A review


Call for Papers Engineering Journal, May 2019

Download Full-Text PDF Cite this Publication

Text Only Version

Computer aided diagnosis of Alzheimer’s disease: A review

  1. Monisha Bharathi

    Dept. of Computer Applications Bharathiar University Coimbatore, India moniarul90@gmail.com

    R. Rajeswari

    Dept. of Computer Applications Bharathiar University Coimbatore, India rrajeswari@rediff.com

    Abstract Alzheimers disease is a neurodegenerative disease which causes memory loss and cognitive decline. Early detection of this disease is essential for effective treatment. Computer aided diagnosis is a tool which is based on medical imaging. It is used to output to assist radiologists in the interpretation of images by improving the accuracy and consistency of diagnosis. It is also used to reduce the image reading time. It is very helpful for physicians to detect the Alzheimers disease. In this paper, various computer aided diagnosis methods which are used in the diagnosis of Alzheimers diseases are examined. This paper provides a detailed overview of the methods used in the diagnosis of Alzheimers disease.

    KeywordsComputer Aided Diagnosis, Alzheimer's Disease, Feature Extraction, Feature Selection, Classification.

    1. INTRODUCTION

      Alzheimers disease (AD) is the most common cause of dementia. AD is a progressive neurodegenerative disorder first affecting memory functions and then gradually affecting all cognitive functions with behavioral impairment and eventually causing death [13, 16]. It has affected more than thirty million people worldwide. It is expected to affect sixty million people over the next 50 years due to the increase in life expectancy and aging of the population. There is no cure for Alzheimers disease, but its early detection is essential for an effective treatment [5].

      Many researchers have developed Computer Aided Diagnosis (CAD) systems that automatically diagnose AD before being detected by cognitive tests [1, 6]. CAD is more important in medical and clinical research. CAD is used for support decisions and it uses the scanned data from medical imaging. CAD system analyzes and evaluates the radiologist in a short period of time. It helps to improve the accuracy and efficiency of radiologists in medical diagnosis.

      The scanned data of brain are stored in different formats such as MRI, fMRI, PET, SPECT. Magnetic Resonance Imaging (MRI) is used to produce high quality of two or three dimensional brain image structures using magnetic fields and radio waves. Functional Magnetic Resonance Imaging (fMRI) is non-invasive technique which is not involving the intrusion of instruments into the body. It allows indirect measurement of neuronal activity [17].

      Positron emission tomography (PET) is a functional imaging technique which produces a three dimensional image of functional processes in the body. PET scanning is used to show blood flow, oxygen and glucose metabolism in the tissues

      of the working brain [3, 12]. Single-photon emission computed tomography (SPECT) acts like PET. It produces images that show how organs work and how blood flows to the heart. It shows the areas of brain which are active and which are less active [3, 11].

      Diagnosis of Alzheimers disease can be implemented by three stages which are: (a) Feature extraction, (b) Feature Selection and (c) Classification. These three stages have several algorithms that can be used for diagnosis. Feature Extraction involves in simplifying the large set of data accurately. Feature Extraction methods are transformative which reduces the dimensionality and describes the data with sufficient accuracy. Feature selection is the process of selecting a subset of relevant features. Feature selection techniques avoid many redundant or irrelevant features. Classification is the most important stage in CAD of AD. It can be categorized as (i) Supervised Classification and (ii) Unsupervised Classification. Supervised Classification is based on trained data. Unsupervised Classification does not make use of trained data.

      The rest of this paper is organized as follows. Section 2 describes the main stages of CAD for AD. Section 3 describes various computer aided diagnosis techniques for Alzheimers disease. Section 4 gives the conclusion.

    2. MAIN STAGES OF CAD FOR AD

      CAD for Alzheimers disease consists of three stages: (a) Feature extraction, (b) Feature Selection and (c) Classification. Various techniques used for these three stages of CAD for AD are shown in figure 1. These stages play a vital role in efficient and early diagnosis of AD.

      Feature extraction is a general term for the methods used to construct the combinations of variables and describing the data with sufficient accuracy. Some of these methods are described here. Partial Least Square (PLS) is a method used for finding the fundamental relations between two matrices. It creates an orthogonal score vector which is also called latent vectors or components by maximizing the covariance between different set of variables [1]. Principal Component Analysis (PCA) is a non- parametric method used for extracting relevant information from confusing data sets. It generates an orthonormal basis vector which can maximize the scatter of all the projected samples [3]. Independent Component Analysis (ICA) is a computational method which is used for illuminating hidden factors that underlie sets of random variables. Lattice Independent Component Analysis (LICA) is a non-linear

      alternative to ICA. It is based on the lattice independent discovered when dealing with noise robustness [4]. CONNectivity matrix linkage (CONN) is a hierarchical agglomerative clustering method which is used to capture information about the neighbourhood relationships in the input space to build clusters [16]. Factor Analysis is a method of statistical techniques which is used to represent a set of variables in terms of hypothetical smaller variables.

      Fig. 1. Algorithms/Methods used for diagnosis

      Feature selection, also known as variable or attribute or variable subset selection, is used to select the subset of relevant features. Fisher Discriminant Ratio (FDR) selects the most discriminant voxels and thus reduces the computational burden in the preliminary voxel selection stage. Linear Discriminant Analysis (LDA) is a method used for separating or characterizing two or more classes of objects or events. Association Rule (AR) is the most popular method which is used for discovering hidden relations between variables in the databases. The Expectation Maximization (EM) is a parameter estimation method which can fall into the general framework of maximum likelihood estimation. It is applied to the data which is considered to be incomplete, or hidden [11]. Mann WhitneyWilcoxon U-Test (MWW) is a novel based method used for a statistical test which can be helpful in the selection task [3].

      Support Vector Machines (SVM) is a supervised learning method mainly used for classification. It can analyze data and recognize patterns [5]. The Random Forest (RF) is an ensemble classification method which is based on learning algorithms. It is used to construct a set of classifiers. It consists of many decision trees [17]. An artificial neural network (ANN) is a paradigm used for information processing which is inspiring by the way of biological nervous systems such as brain [3].

    3. OVERVIEW OF VARIOUS CAD TECHNIQUES FOR AD Segovia et al [1] have proposed an early diagnosis of

      Alzheimers disease based on Partial Least Square algorithm for extraction of score vectors and selecting a number of scores as features. Finally diagnosis is done by Support Vector Machine (SVM) classifier. The authors have used 9 SPECT images for normal and AD patient. They have used 41 normal SPECT and 56 AD images from the image database.

      Andrea Chincarini et al [2] have proposed local MRI analysis approach in the diagnosis of early prodromal Alzheimers disease. The authors have used intensity and textural for features and template matching techniques for feature extraction of voxel of interest. SVM is used for classification.

      Lopez et al [3] have proposed Principal Component Analysis (PCA) based techniques and supervised classification schemes for the early detection of Alzheimers disease. Principal Component Discriminant methods are proposed as feature extraction and Linear Discriminant Analysis (LDA) or the Fisher Discriminant Ratio (FDR) is used for feature selection. Finally the classification is done by SVM classifier and Neural Network Classifier. The authors have used 53 AD images, 114 Mild Cognitive Impairment (MCI) images and 52 Normal Control images for their experiment.

      Darya Chyzhyk et al [4] have proposed a Hybrid Dendritic computing with kernel- LICA applied to Alzheimers disease detection in MRI. Here the authors have proposed Lattice Independent Component Analysis (LICA) and kernel Approach for feature extraction and SVM for Classification.

      Illan et al [5] have proposed Bilateral symmetry aspects in Computer-aided Alzheimers disease diagnosis by single- photon emission-computed tomography imaging. The authors have studied the brain symmetry and its usage for CAD of Alzheimers disease. They have studied the symmetry of brain using machine learning based classifiers and have studied their relationship with AD. For this purpose they have used the eigen image decomposition of single-photon emission computed tomography images. The authors have used 97 images, out of which 41 are images of normal patient, 30 are perfusion deficit (AD1) images, 22 are moderate deficit (AD2) images and 4 are severe deficit (AD3) images.

      Martinez-Murcia et al [6] have proposed Computer-aided diagnosis for Alzheimers disease based on Mann-Whitney- Wilcoxon U-Test for selecting a voxel and factor analysis as feature extraction. Finally diagnosis is done by SVM classifier. Here the authors have used 96 SPECT images and 196 PET images.

      Chaves et al [7] have proposed an association rule (AR)- based feature selection method for AD diagnosis. This rule based feature selection had enabled in solving the sample size problem in order to design a CAD system. They have proposed PCA or PLS for feature extraction and SVM as classifier. The authors have used 150 PET images, out of which 75 are control patient images and 75 are AD patient images.

      Chaves et al [8] have proposed integrating discretization and association rule based classification for Alzheimers disease diagnosis. The discretization method is used as feature extraction for selecting the mean intensity. Association rule method is used for classification.

      Alexandre Savio et al [9] have proposed deformation based feature selection for Computer Aided Diagnosis of Alzheimers disease. In this paper the authors have used the scalar deformation measures of CAD systems for AD. They have evaluated three supervised feature selection methods. They are Pearsons Correlation, Bhattacharyya distance, Welchs t-test. Final stage of diagnosis is performed by SVM classifier.

      TABLE I. OVERVIEW OF VARIOUS METHODS ASSOCIATED WITH CAD FOR AD

      Methods

      Features

      Feature Extraction

      Feature Selection

      Classification

      Segovia et al

      Out-of-bag error Score vector

      Partial Least Squares

      Fisher Discriminant Ratio

      SVM

      Andrea Chincarini et al

      intensity and textural MRI-based features

      template matching techniques

      SVM

      Random Forest

      Lopez et al

      Voxel Selection

      Principal Component Analysis

      Fisher Discriminant Ratio

      (FDR)

      SVM

      Linear Discriminant Analysis (LDR)

      Neural Network

      Darya Chyzhyk et al

      Lattice Independent Component Analysis

      SVM

      Kernel Approaches

      Illan et al

      Eigen vector

      Principal Component

      Analysis Symmetric

      SVM

      Martinez-Murcia et al

      Voxel Selection

      Factor Analysis

      MannWhitneyWilcoxon U-

      Test (MWW)

      SVM

      Chaves et al

      Voxel Selection

      Principal Component Analysis

      Partial Least Squares

      Association Rule

      SVM

      Chaves et al

      Mean Intensity

      Discretization Method

      AR

      Alexandre Savio et al

      Scalar Measures

      Jacobian Determinant

      SVM

      Salas-Gonzalez et al

      Mean Intensity

      Mean Image Standard Deviation Image

      Quadratic Discriminant Linear discriminant function Mahalanobis distance

      SVM

      K nearest neighbour

      Gorriz et al

      Voxel intensities

      Gaussian mixture model

      EM

      SVM

      Illan et al

      Voxel as Features

      Principal Component Analysis

      Kernel-SVM

      Independent Component

      Analysis

      Ramirez et al

      Voxel Intensities

      Sequential Minimal Optimization (SMO)

      FDR

      SVM

      Segovia et al

      Feature Vectors

      Gaussian mixture model

      SVM

      Score Vectors

      Partial Least Squares

      Andres Ortiz et al

      Voxel Intensities

      CONNectivity matrix linkage

      clustering Algorithm and LVQ3

      Fisher Discriminant Ratio (FDR)

      SVM

      Evanthia et al

      Demographics Behavioral Head motion

      Volumetric measures Activation patterns

      Hemodynamics

      Wrapper approach Filter approach

      SVM

      Random Forest

      Illán et al

      Voxel selection

      Component Based Analysis

      Image Factorization

      SVM

      Alvarez Illan et al

      Independent Component

      Analysis (ICA)

      Kernel-SVM

      Chaves et al

      Voxel selection

      Normalized Mean Square Error (NMSE)

      t-test

      SVM

      Evanthia et al

      Demographics Behavioral Head motion

      Volumetric measures Activation patterns

      Hemodynamics

      Symmetrical Uncertainty

      Random Forest

      1. Pearsons correlation

      2. Bhattacharyya distance (BD)

      3. Welchs t-test (WT)

      Salas-Gonzalez et al [10] have proposed two approaches to select set of voxels for the diagnosis of Alzheimers disease. The first approach used by authors is based on selecting the voxels which have the greatest difference between controls and Alzheimers disease. The second approach is based on selecting the voxels which not only have greatest difference between both modlities but also present lower dispersion. The authors have used 41 normal images, 20 possible AD images, 17 probable AD images, 1 certain AD images.

      Gorriz et al [11] have proposed GMM based SPECT image classification for the diagnosis of Alzheimers disease. The authors have used voxel intensities as features and Gaussian mixture model for feature extraction. The Expectation Maximization Algorithm is used as feature selection based on Gausssian result. Classification is done by SVM classifier. The authors have used 97 SPECT images, out of which 43 are normal images, 30 are possible AD images, 20 are probable AD images and 4 are certain AD images.

      Illan et al [12] have proposed F-FDG PET imaging analysis for Computer aided Alzheimers diagnosis. Here the authors have proposed PCA/ICA as feature extraction and voxel as features. SVM is used for performing the classification task. The authors have used 97 normal control images, 209 Mild Cognitive Impairment(MCI) images and 95 AD images.

      Ramirez et al [13] have proposed Computer-aided diagnosis of Alzheimers type dementia combining support vector machines and discriminant set of features. The authors have used Fisher Discriminant Ratio as feature selection and have used SVM as classifier. The authors have used 52 SPECT images of Patients. In that 23 are Normal control images, 13 are possible AD images, 12 are probable AD images and 4 are certain AD images.

      Segovia et al [14] have proposed a comparative study of feature extraction methods for the diagnosis of Alzheimers disease using the ADNI database. In this paper the authors have proposed two approaches. The first approach is to select the feature vectors and Gaussian mixture model for feature extraction. The second approach is to select the score vector and Partial Least Square method for feature extraction. Finally SVM is used for classification. The authors have used PET images for experiment results. In that they used 97 normal control images, 188 Mild Cognitive Impairment (MCI) images, 23 MCI converter images and 95 AD images.

      Evanthia et al [15] have proposed a six stage approach for the diagnosis of the Alzheimers disease based on fMRI data. The authors have categorized feature extraction into six stages. They are demographics, behavioral, head motion, volumetric measures, activation patterns, hemodynamics. Using these they have extracted the features and have used wrapper approach and filter approach for feature selection. SVM and Random forest are used for classification. The authors have also proposed a supervised method to assist the diagnosis and monitor progression of Alzheimers disease using data from an fMRI experiment [17]. Here the authors have used the feature extraction methods which they have proposed in [15]. They have used symmetric uncertainty for feature selection and random forest for classification [17].

      Andres Ortiz et al [16] have proposed Learning Vector Quantization(LVQ)-SVM based CAD tool applied to structural MRI for the diagnosis of the Alzheimers disease. In this paper

      the authors proposed CONNectivity matrix linkage clustering Algorithm as feature extraction for clustering. LVQ is used for reduction of features and Fisher discriminant for feature selection. The classification is done by SVM. Here the authors have used 50 T1 weighted MRI images, out of which 25 are normal images and 25 are AD images.

      Illan et al [18] has proposed Computer aided diagnosis of Alzheimers disease using component based SVM. The image factorization is done by dividing the whole brain image into smaller subvolumes or components. The classification using SVM is done for each component. The authors have used 79 SPECT images of Patients. In that 41 are Normal control images, 20 are possible AD images, 14 are probable AD images and 4 are certain AD images.

      Alvarez Illan et al [19] have proposed projecting independent components of SPECT images for computer aided diagnosis of Alzheimers disease. The authors have used Independent Component Analysis for feature extraction and to reduce the feature space dimensionality. Classification is done by SVM. The authors have used 79 SPECT images of Patients. Out of these images 41 are Normal control images, 20 are possible AD images, 14 are probable AD images and 4 are certain AD images.

      Chaves et al [20] have proposed SVM-based computer- aided diagnosis of the Alzheimers disease using t-test NMSE feature selection with feature correlation weighting. Classification is done by SVM. Here the authors have used 79 SPECT images, out of which 41 are images of normal patients and 38 are images of AD affected patients.

    4. CONCLUSION

CAD tools help in early diagnosis of AD and help in slowing down the rapid progress of the disease. The main stages used in the CAD of AD are: feature extraction, feature selection and classification. In this survey, various algorithms and methods which are used in these stages of diagnosis are studied in detail. For every method a brief discussion and analysis are presented. This survey would enable the beginners in this research area to get an overview of the various techniques used in CAD of AD.

ACKNOWLEDGEMENT

The authors are thankful to Bharathiar University for valuable support.

REFERENCES

  1. F. Segovia , J.M. Gorriz, J. Ramirez, D. Salas-Gonzalez, I. Alvarez, Early diagnosis of Alzheimers disease based on Partial Least Squares and Support Vector Machine, Expert Systems with Applications 40 (2013), pp. 677-683.

  2. Andrea Chincarini, Paolo Bosco, Piero Calvini, Gianluca Gemme, Mario Esposito, Chiara Olivieri, Luca Rei, Sandro Squarcia, Guido Rodriguez, Roberto Bellotti, Piergiorgio Cerello, Ivan De Mitri, Alessandra Retico, Flavio Nobili, Local MRI analysis approach in the diagnosis of early and prodromal Alzheimers disease, NeuroImage 58 (2011), pp. 469- 480.

  3. M. Lopez, J. Ramirez, J.M. Gorriz, I. Alvarez, D. Salas-Gonzalez, F. Segovia, R. Chaves, P. Padilla, M. Gomez-Rio, the Alzheimers Disease

    Neuroimaging Initiative, Principal component analysis-based techniques and supervised classification schemes for the early detection of Alzheimers disease, Neurocomputing 74 (2011), pp. 1260-1271.

  4. Darya Chyzhyk, Manuel Grana , Alexandre Savio, Josu Maiora, Hybrid dendritic computing with kernel-LICA applied to Alzheimers disease detection in MRI, Neurocomputing 75 (2012), pp. 72-77.

  5. Ignacio Alvarez Illan, Juan Manuel Gorriza, Javier Ramirez, Elmar W. Lang, Diego Salas-Gonzalez, Carlos G. Puntonet, Bilateral symmetry aspects in computer-aided Alzheimers disease diagnosis by singlephoton emission-computed tomography imaging, Artificial Intelligence in Medicine 56 (2012), pp. 191-198.

  6. F.J. Martinez-Murcia, J.M. Gorriz, J. Ramirez, C.G. Puntonet, D. Salas- Gonzalez, Computer Aided Diagnosis tool for Alzheimers Disease based on Mann-Whitney-Wilcoxon U-Test, Expert Systems with Applications 39 (2012), pp. 9676-9685.

  7. R. Chaves, J. Ramirez, J.M. Gorriz, C.G. Puntonet, Association rule based feature selection method for Alzheimers disease diagnosis, Expert Systems with Applications 39 (2012), pp. 11766-11774.

  8. R. Chaves, J. Ramirez, J.M. Gorriz, Integrating discretization and association rule-based classification for Alzheimers disease diagnosis, Expert Systems with Applications 40 (2013), pp. 1571-1578.

  9. Alexandre Savio, Manuel Grana, Deformation based feature selection for Computer Aided Diagnosis of Alzheimers Disease, Expert Systems with Applications 40 (2013), pp. 1619-1628.

  10. D. Salas-Gonzalez, J.M. Gorriza, J. Ramirez, I. Alvarez, M. Lopez, F. Segovia , C.G. Puntonet, Two approaches to selecting set of voxels for the diagnosis of Alzheimers disease using brain SPECT images, Digital Signal Processing 21 (2011), pp. 746-755.

  11. J.M. Gorriz, F. Segovia, J. Ramirez, A. Lassl, D. Salas-Gonzalez, GMM based SPECT image classification forthe diagnosis of Alzheimers disease, Applied Soft Computing 11 (2011), pp. 2313- 2325.

  12. I.A. Illan, J.M. Gorriz, J. Ramirez, D. Salas-Gonzalez, M.M. Lopez, F. Segovia, R. Chaves, M. Gomez-Rio, C.G. Puntonet, F-FDG PET

    imaging analysis for computer aided Alzheimers diagnosis,

    Information Sciences 181 (2011), pp. 903-916.

  13. J. Ramirez, J.M. Gorriz, D. Salas-Gonzalez, A. Romero, M. Lopez, I. Alvarez, M.Gomez-Rio, Computer-aided diagnosis ofAlzheimers type dementia combining support vector machines and discriminant set of features, Information Sciences 237 (2013), pp. 59-72.

  14. F. Segovia, J.M. Gorriz, J. Ramirez, D. Salas-Gonzalez, I. Alvarez, M. Lopez, R. Chaves, A comparative study of feature extraction methods for the diagnosis of Alzheimers disease using the ADNI database, Neurocomputing 75 (2012), pp. 64-71.

  15. Evanthia E. Tripoliti, Dimitrios I. Fotiadis, Maria Argyropoulou, George Manis, A six stage approach for the diagnosis of the Alzheimers disease based on fMRI data, Journal of Biomedical Informatics 43 (2010), pp. 307-320.

  16. Andres Ortiz, Juan M. Gorriz, Javier Ramirez , F.J. Martinez Murcia, LVQ-SVM based CAD tool applied to structural MRI for the diagnosis of the Alzheimers disease, Pattern Recognition Letters 34 (2013), pp. 1725-1733.

  17. Evanthia E. Tripoliti, Dimitrios I. Fotiadis, Maria Argyropoulou, A supervised method to assist the diagnosis and monitor progression of Alzheimers disease using data from an fMRI experiment, Artificial Intelligence in Medicine 53 (2011), pp. 35- 45.

  18. I.A. Illan, J.M. Gorriza, M.M. Lopez, J. Ramirez, D. Salas-Gonzalez, F. Segovia, R. Chaves, C.G. Puntonet, Computer aided diagnosis of Alzheimers disease using component based SVM, Applied Soft Computing 11 (2011), pp. 2376-2382.

  19. I. Alvarez Illan, J.M. Gorriz, J. Ramirez, D. Salas-Gonzalez, M. Lopez,

    F. Segovia, P. Padilla, C.G. Puntonet, Projecting independent components of SPECT images for computer aided diagnosis of Alzheimers disease, Pattern Recognition Letters 31 (2010), pp. 1342- 1347.

  20. R. Chaves, J. Ramirez, J.M. Gorriz, M. Lopez, D. Salas-Gonzalez, I. Alvarez, F. Segovia, SVM-based computer-aided diagnosis of the Alzheimers disease using t-test NMSE feature selection with feature correlation weighting, Neuroscience Letters (2009).

Leave a Reply

Your email address will not be published. Required fields are marked *