Application of fuzzy logic for Alzheimer’s disease diagnosis Igor Krashenyi, Anton Popov
Javier Ramirez, Juan Manuel Gorriz
Physical and Biomedical Electronics Department National Technical University of Ukraine “Kyiv Polytechnic Institute” Kyiv, Ukraine
[email protected] [email protected]
Department of Signal Theory, Telematics and Communications University of Granada Granada, Spain
[email protected] [email protected]
Abstract— Fuzzy Inference System (FIS) is developed using subtractive clustering algorithm, and applied to classification between MRI images of patients having Mild Cognitive Impairment (MCI) or Alzheimer’s Disease (AD) and Normal Controls (NC). Features used as FIS inputs are mean values and standard deviations in intensities from most descriptive brain regions. k-fold cross-validation was used to estimate FIS performance, resulting in accuracy, sensitivity, specificity and positive predictive value (ppv) characteristics of FIS classification between different groups. ppv was equal to 0.8778±0.0088 (AD vs. NC), 0.7289±0.0243 (NC vs. MCI), and 0.8531±0.0069 (MCI vs. AD).
for classification of AD presence, the features of MRI images should be defined. Previously we proposed to calculate features for different brain regions with selection of the most prone to tissue changes in AD and to use separated FIS for every type of features [14]. In this paper we continue the study on possibility to use Fuzzy Inference for early detection of AD and Mild Cognitive Impairment with different types of features. The aim of present study is to develop the FIS using means and standard deviations of MRI voxel intensity and estimate its performance.
Keywords—Alzheimer's disease, MRI, classification, mildcognitive impairment, fuzzy logic
I. INTRODUCTION Alzheimer's is a type of dementia that leads to problems with memory, thinking and behavior. Alzheimer's is the most common form of dementia, a general term for memory loss and other intellectual abilities serious enough to interfere with daily life. AD has no cure, but treatments to reduce symptoms are available and research continues. To reduce symptoms and to stanch the disease early diagnosis has a huge importance. Magnetic resonance imaging (MRI) is the one of the most conventional tools for objective AD diagnosis [2]. There are many different types of classification methods for automatic diagnosis of AD, such as support vector machines [3],[4],[5],[6], Bayes classifiers [6], neural networks [8],[9],[10], random forests [11],[12],[13], but none of them provides robust information about the stage of the AD, they can just indicate presence of disease. AD is a very slow and continues process [1] and borders between stages are very defuse. That is why it is proposed to use fuzzy inference system as classifier for AD diagnosis [14]. A fuzzy inference system (FIS) is a system that uses fuzzy set theory to map inputs (features) to outputs (classes) [17],[18]. Fuzzy inference is the process of formulating the mapping from a given input to an output using fuzzy logic. To use FIS
Data used in preparation of this article were obtained from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) database (adni.loni.usc.edu). As such, the investigators within the ADNI contributed to the design and implementation of ADNI and/or provided data but did not participate in analysis or writing of this report. A complete listing of ADNI investigators can be found at: http://adni.loni.usc.edu/wpcontent/uploads/how_to_apply/ADNI_Acknowledgement_List.pdf
II. MATERIALS AND METHODS A. Database Data used in the preparation of this article were obtained from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) database (adni.loni.usc.edu). ADNI database contains 1.5T and 3.0T MRI scans. A database consisting of 818 subjects (229 normal control (NC), 401 mild cognitive impairment (MCI) and 188 AD subjects) was used in this study to train, validate and test the FIS for detecting AD. Demographic data of patients is presented in Table 1. TABLE I.
DEMOGRAPHIC DATA OF PATIENTS IN ADNI DATABASE
Diagnosis
Number
Age
NC MCI AD
229 401 188
75.97±5.0 74.85±7.4 75.36±7.5
Gender (M/F) 119/110 258/143 99/98
MMSE 29.00±1.0 27.01±1.8 23.28±2.0
B. Pre-processing Image data pre-processing, segmentation and coregistration of T1-weighted MRI images from the ADNI database have been performed. Initially, images from the ADNI database were nor skull-stripped neither spatially normalized. Thus, all the images had to be pre-processed and co-registered before segmentation. The whole process has
Fig. 1. FIS internal structure.
been performed using the voxel-based morphometry (VBM) toolbox [19] for statistical parametric mapping (SPM) [20]. C. Feature extraction Feature extraction was performed using special atlas of brain regions [21]. The atlas was co-registered preserving the ROI labels to the spatial normalization template that matched the MRI resolution. Each voxel is an integer number in a range from 1 to 116 that matches to one of 116 ROIs. The p-value of t-test was used for ROIs selection purposes [14]. Twenty-four the most discriminant ROIs were selected for feature extraction purposes. Over every of 24 ROIs mean values (1) and standard deviations (2) were calculated: n
μ=
1 ∑x , n i =1 i
(1)
n
σ=
2 1 ( xi − μ ) , ∑ n i =1
(2)
where x i – voxel intensity; n – number of voxels in a single ROI. D. Fuzzy inference system for AD diagnosis Fuzzy logic is a mathematical approach for computing and inferencing that generalizes classical logic and set theory employing the concept of fuzzy set. The process of fuzzy inference involves all of the pieces that are described in membership functions, logical operations, and if-then rules [18].
A fuzzy inference system (FIS) is a system that uses fuzzy set theory to map inputs (features) to outputs (classes). To construct FIS, first one needs to select the input numerical variables, which should be crisp, and define their ranges for every terms. Then during fuzzyfication stage, the correspondence between the input values and every fuzzy set should be defined. This is done with the help of membership functions, which represent the degree of membership of the parameter value to each class. After that, the set of fuzzy rules that describe decisions of FIS using logical operators and/or should be set, and the rule for combining fuzzy outputs from each rule should be defined. Finally, the output distribution from combination of fuzzy rules should be obtained, followed by defuzzyfication to get the crisp classification result. In current work, it is proposed to use mean values and standard deviations of MRI intensity over 24 most discriminant ROIs as input features [14]. FIS consists of fuzzification, defuzzification inference and rule parts, as it is shown in Fig. 1. Proposed FIS was constructed using subtractive clustering algorithm [16]. In Fig. 2 the internal structure of proposed approach is given. III. EXPERIMENTAL RESULTS In this study data from 818 patients were analyzed. To measure the performance of the proposed classification system, several statistical parameters [22] are widely used. Accuracy (acc), sensitivity (sens), positive predictive value (ppv) and specificity (spc) rates were calculated at each iteration of k-fold cross-validation as follows:
TP + TN , TP + TN + FN + FP TP , sens = TP + FN TP , ppv = TP + FP TN , spc = TN + FP
acc =
(3)
Fig. 2. Internal structure of proposed classification system.
Computer-aided diagnosis (CAD) systems based on support vector machines and random forests classifiers in combination with feature extraction techniques based on partial least squares and principal components are showing the best performance. However, these methods allow only finding presence of disease in a crisp manner, i.e. patient has AD or not. But it is known that AD is a very slow-progressing and protracted disease, and there is no clear differentiation between its stages. This results in the indecipherable boundaries between the normal state and AD state. The most significant merit of proposed approach is that the output of such system shows the membership of patient to Normal, MCI or AD class. As we see in experiments, our approach results in fairly good classification quality in terms of specificity, sensitivity, accuracy and positive predictive value, especcialy for AD vs. Normal. It is promising to use different types of features to improve performance of proposed FIS for AD diagnosis. Proposed fuzzy approach for AD diagnosis gives a possibility to use Mini Mental State Examination (MMSE) as labels for patients in training set.MMSE is the most commonly used instrument for screening cognitive function.
acc sens spc ppv
Normal vs. MCI
DISCUSSION
EXPERIMENTAL RESULTS
MCI vs. AD
TABLE II.
AD vs. Normal
where TP – number of correctly classified AD patients; TN – number of NC patients classified as AD; FN – number of AD patients classified as NC; TN – number of correctly classified NC patients. In current approach it is proposed to measure classifier performance by k-fold cross-validation technique [23]. Crossvalidation consists in averaging several validations relied on a single split of data estimators of the risk corresponding to different data splits. To perform k-fold validation, datasets with all patients is randomly divided into k parts of same size. k-fold cross-validation will be done k times. At each stage, one fold gets to play a role of testing set whereas the other remaining k-1 parts are used as training set. This procedure must be repeated afterwards. During this research, three experiments were carried out. The first one was dedicated to binary classification of AD and Normal subjects; the second one was dedicated to binary classification of Normal; and MCI subjects, and the third one was dedicated to binary classification of MCI and AD subjects. Results of these experiments are given in Tables 2.
0.8811±0.0031 0.8605±0.0065 0.9106±0.0044 0.8778±0.0088
0.7283±0.0057 0.7152±0.0148 0.7100±0.0076 0.8531±0.0069
0.7272±0.0139 0.9375±0.0043 0.7713±0.0169 0.7289±0.0243
This examination is not suitable for making a diagnosis but can be used to indicate the presence of cognitive impairment. This could improve output performance and give a possibility to determine stage more accurate instead of class belongings. However, proposed technique seems to be promising because of possibility to describe in flexible manner patients’ labels and absence of crisp borders between stages. CONCLUSION Fuzzy inference classification system presented in this work, uses segmented GM MRI images from ADNI database. Means and standard deviations of MRI intensities over the most informative ROIs are used as input features for fuzzy inference system. Overall performance that was estimated using a k-fold cross-validation methodology is characterized by 86% sensitivity, 91% specificity, 88% accuracy and 88% positive predictive value for AD vs. Normal classification.
ACKNOWLEDGMENT Data collection and sharing for this project was funded by the Alzheimer’s Disease Neuroimaging Initiative (ADNI) (National Institutes of Health Grant U01 AG024904) and DOD ADNI (Department of Defense award number W81XWH-12-2-0012). ADNI is funded by the National Institute on Aging, the National Institute of Biomedical Imaging and Bioengineering, and through generous contributions. ADNI data are disseminated by the Laboratory for Neuroimaging at the University of Southern California. This work was partly supported by the MINECO under the TEC2012–34306 project and the Consejería de Innovación, Ciencia y Empresa (Junta de Andalucía, Spain) under the Excellence Projects P09–TIC–4530 and P11–TIC–7103. Igor Krashenyi was supported by Erasmus-Mundus EMERGE scholarship for PhD students during the work on this research. Check http://emerge.uaic.ro/ for extra information.
[9]
[10]
[11]
[12]
[13]
[14]
REFERENCES [1] [2] [3]
[4]
[5]
[6]
[7]
[8]
Mayeux, R. (2010). Early Alzheimer’s disease. New England Journal of Medicine 362, 2194–2201. Vemuri, P., and Jack Jr, C.R. (2010). Role of structural MRI in Alzheimer’s disease. Alzheimers Res Ther 2, 23. Duchesne, S., Caroli, A., Geroldi, C., Barillot, C., Frisoni, G.B., and Collins, D.L. (2008). MRI-Based Automated Computer Classification of Probable AD Versus Normal Controls. IEEE Transactions on Medical Imaging 27, 509–520. Álvarez, I., Górriz, J.M., Ramírez, J., Salas-Gonzalez, D., López, M., Segovia, F., Puntonet, C.G., and Prieto, B. (2009). Alzheimer’s diagnosis using eigenbrains and support vector machines. In Bio-Inspired Systems: Computational and Ambient Intelligence, (Springer), pp. 973–980. Górriz, J.M., Ramírez, J., Lassl, A., Salas-Gonzalez, D., Lang, E.W., Puntonet, C.G., Álvarez, I., López, M., and Gómez-Río, M. (2008). Automatic computer aided diagnosis tool using component-based SVM. In Nuclear Science Symposium Conference Record, 2008. NSS’08. IEEE, (IEEE), pp. 4392–4395. Morra, J.H., Zhuowen Tu, Apostolova, L.G., Green, A.E., Toga, A.W., and Thompson, P.M. (2010). Comparison of AdaBoost and Support Vector Machines for Detecting Alzheimer’s Disease Through Automated Hippocampal Segmentation. IEEE Transactions on Medical Imaging 29, 30–43. López, M., Ramírez, J., Górriz, J.M., Salas-Gonzalez, D., Alvarez, I., Segovia, F., and Puntonet, C.G. (2009). Automatic tool for Alzheimer’s disease diagnosis using PCA and Bayesian classification rules. Electronics Letters 45, 389–391. Huang, C., Yan, B., Jiang, H., and Wang, D. (2008). Combining Voxelbased Morphometry with Artifical Neural Network Theory in the
[15]
[16]
[17] [18] [19] [20]
[21]
[22] [23]
Application Research of Diagnosing Alzheimer’s Disease. (IEEE), pp. 250–254. Padilla, P., Górriz, J.M., Ramirez, J., Chaves, R., Segovia, F., Alvarez, I., Salas-González, D., López, M., and Puntonet, C.G. (2010). Alzheimer’s disease detection in functional images using 2D Gabor wavelet analysis. Electronics Letters 46, 556–558. Zhang, Y., Dong, Z., Wu, L., and Wang, S. (2011). A hybrid method for MRI brain image classification. Expert Systems with Applications 38, 10049–10053. Tripoliti, E.E., Fotiadis, D.I., and Manis, G. (2012). Automated Diagnosis of Diseases Based on Classification: Dynamic Determination of the Number of Trees in Random Forests Algorithm. IEEE Transactions on Information Technology in Biomedicine 16, 615–622. Ramirez, J., Chaves, R., Gorriz, J.M., Lopez, M., Lvarez, I.A., SalasGonzalez, D., Segovia, F., and Padilla, P. (2009). Computer aided diagnosis of the Alzheimer’s disease combining spect-based feature selection and random forest classifiers. In Nuclear Science Symposium Conference Record (NSS/MIC), 2009 IEEE, (IEEE), pp. 2738–2742. Ramírez, J., Górriz, J.M., Segovia, F., Chaves, R., Salas-Gonzalez, D., López, M., Álvarez, I., and Padilla, P. (2010). Computer aided diagnosis system for the Alzheimer’s disease based on partial least squares and random forest SPECT image classification. Neuroscience Letters 472, 99–103. Krashenyi, I., Ramírez, J., Popov, A., Górriz, J.M. (2015). Fuzzy inference system for Alzheimer’s disease diagnosis. Current Alzheimer Research [submitted for publication]. Villain, N., Desgranges, B., Viader, F., de la Sayette, V., Mezenge, F., Landeau, B., Baron, J.-C., Eustache, F., and Chetelat, G. (2008). Relationships between Hippocampal Atrophy, White Matter Disruption, and Gray Matter Hypometabolism in Alzheimer’s Disease. Journal of Neuroscience 28, 6174–6181. Bataineh, K.M., Naji, M., and Saqer, M. (2011). A comparison study between various fuzzy clustering algorithms. Jordan Journal of Mechanical and Industrial Engineering 5, 335–343. Zadeh, L.A. (1965). Fuzzy sets. Information and Control 8, 338–353., Zadeh, L.A. (1968). Fuzzy algorithms. Information and Control 12, 94– 102. Ashburner, J., and Friston, K.J. (2000). Voxel-Based Morphometry—The Methods. NeuroImage 11, 805–821. Friston, K.J., Holmes, A.P., Worsley, K.J., Poline, J.-P., Frith, C.D., and Frackowiak, R.S. (1994). Statistical parametric maps in functional imaging: a general linear approach. Human Brain Mapping 2, 189–210. Alemán-Gómez, Y., Melie-García, L., Valdés-Hernandez, P. (2006) IBASPM: Toolbox for automatic parcellation of brain structures. 12th Annual Meeting of the Organization for Human Brain Mapping (CDRom in NeuroImage), Vol. 27, No.1. Rice, J.A. (2007). Mathematical statistics and data analysis (Belmont, CA: Thomson/Brooks/Cole). Arlot, S., and Celisse, A. (2010). A survey of cross-validation procedures for model selection. Statistics Surveys 4, 40–79.