Risk Assessment of Alzheimer’s Disease using the Information Diffusion Model from Structural Magnetic Resonance Imaging

Abstract

Background: Recently, automatic risk assessment methods have been a target for the detection of Alzheimer’s disease (AD) risk.

Objective: This study aims to develop an automatic computer-aided AD diagnosis technique for risk assessment of AD using information diffusion theory.

Methods: Information diffusion is a fuzzy mathematics logic of set-value that is used for risk assessment of natural phenomena, which attaches fuzziness (uncertainty) and incompleteness. Data were obtained from voxel-based morphometry analysis of structural magnetic resonance imaging.

Results and Conclusion: The information diffusion model results revealed that the risk of AD increases with a reduction of the normalized gray matter ratio (p > 0.5, normalized gray matter ratio <40%). The information diffusion model results were evaluated by calculation of the correlation of two traditional risk assessments of AD, the Mini-Mental State Examination and the Clinical Dementia Rating. The correlation results revealed that the information diffusion model findings were in line with Mini-Mental State Examination and Clinical Dementia Rating results. Application of information diffusion model contributes to the computerization of risk assessment of AD, which has a practical implication for the early detection of AD.

Keywords

Alzheimer’s disease computer-aided AD diagnosis early detection gray matter volume information diffusion theory risk assessment

INTRODUCTION

In older adults, Alzheimer’s disease (AD) is the most common cause of dementia, associated with volume loss of the whole brain and changes in the region of the brain that are responsible for memory,learning, and higher executive functioning [1]. It is estimated that 5.3 million Americans of all ages suffer from AD in 2015. This number is expected to increase to 11–16 million people by 2050. AD is the only top 10 cause of death in America that cannot be cured or prevented [1]. Among the several neuroimaging modalities, magnetic resonance imaging (MRI) is more widely used in AD-related studies because of its excellent spatial resolution, high availability, good contrast, and lack of radioactive pharmaceutical injections, which are used in positron emission tomography (PET) and single-photon emission computed tomography (SPECT) [2]. Recently, several studies have been using MRI biomarkers in AD investigations based on structural MRI [3 –10], functional MRI [11], and diffusion tensor imaging [12 –15]. Atrophy measured on structural MRI (s-MRI) is a powerful biomarker of the stage and intensity of the neurodegenerative aspect of AD pathology [16]. s-MRI measures brain morphometry and therefore can capture gray matter atrophy related to the loss of neurons and synapses, and the dendritic de-arborization that occurs on a microscopic level in AD [16]. Researchers in [17] provided physicians with practical guidance on risk assessment of AD. They reported evidence-based guidelines using systematic literature research for genetic risk factors and general risk factors (e.g., education, chemical exposure, occupation, and hypertension). In [18], the authors investigated the dynamic of gray matter loss in AD patients in different parts if brain. In this study, we are only focusing on AD Risk Assessment using overall gray matter (GM) atrophy obtained from s-MRI data. Voxel-based morphometry (VBM), introduced by Ashburner and Friston, has been developed to provide a powerful way to investigate s-MRI scans [3 , 19–21]. VBM is a method used to assess the whole-brain structure with voxel-by-voxel comparisons and has been developed to analyze tissue concentrations or volumes between subject groups to distinguish degenerative diseases with dementia. Moreover, VBM has been reported to be a surrogate indicator of the full-brain topographic representation of the neurodegenerative aspects of AD pathology [2, 20]. The purpose of the current study is to perform risk analyses for AD based on overall structural gray matter atrophy using the Information Diffusion Model (IDM) method plus the VBM technique with small data-sample sizes.

The main characteristic of risk factors is uncertainty [22], because such issues entail unpredicted variables to be either divided or integrated in diverse cases, as the process of granulation is not fixed. One of the main advantages of IDM is the ability it provides for the calculation of risk in uncertain conditions [23]. When the risk is subjected to probability and statistical analysis, the uncertainty that is involved is relevant to the similar degree of standard mode, which can be depicted by fuzzy-sets membership in mathematics theory. Information diffusion is a fuzzy mathematics logic of the set-value method, which evaluates optimizing the use of fuzzy information [24]. It is frequently used to assess exceeding probability of risk factors in various fields, such as medicine and the environment [25, 26].

The diagnosis of AD, as with many other human diseases, suffers from a lack of certainty, and it is important to develop an automated approach to address this uncertainty. The Mini-Mental State Examination (MMSE) [27, 28], and the Clinical Dementia Rating (CDR) [29], are two traditional methods for the estimation of AD risk. These methods are based on simple surveys, which are associated with common method bias [30]. The bias in surveys escalates the fuzziness of the diagnosis of AD.

IDM is a reliable model to mathematically deal with the available sample for an accurate estimation. The reason that IDM is useful in tackling uncertainty is that mathematics changes observations into normal fuzzy sets [31]. Apart from the fuzziness of AD, it is a helpful model for the estimation of the disease when a large sample size is not available [25, 32]. This is plausible, as the extracted data, no matter how small, is sufficient for the proposed method to assess and predict risk.

To the best of our knowledge, this study is the first to apply a mathematical approach (the IDM) in the calculation of AD risk based on GM volume (GMV). The proposed approach can be considered a computer-aided AD diagnosis (CAAD) system. To validate the IDM results, correlations between the findings of the proposed model with the MMSE and CDR have been investigated. These approaches are frequently used for detection of the risk of AD [27 , 33]. In accordance with the literature, it is hypothesized that IDM has adequate performance in assessment of AD risk based on global gray matter atrophy. In summary, the aim of this study was to introduce a novel and automatic CAAD technique for risk assessment of AD using IDT based on global gray matter atrophy of s-MRI, which can be considered for small data-sample sizes.

The rest of this paper is presented as follows: section two provides a description of the subjectschosen for this work. Section three describes the proposed methodology to design an automated, high-performance AD risk assessment system. The experimental results, discussion, and evaluation of the proposed system are given in section four. Finally, section five presents the conclusions.

MATERIALS AND METHODS

MRI data set

The MRI images and data used in this work were obtained from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) database [34].

MRI parameters

All MRI scans in this study were obtained using 3T scanners manufactured by Siemens. The acquisition parameters on the Siemens scanner were: acquisition plane = sagittal, acquisition type = 3D, coil = PA, field strength = 3.0 tesla, flip angle = 9.0 degrees, matrix X = 240.0 pixels, matrix Y = 256.0 pixels, matrix Z = 176, mfg model = Skyra, pixel spacing X = 1.0 mm, pixel spacing Y = 1.0 mm, pulse sequence = GR/IR, slice thickness = 1.2 mm, TE = 2.98 ms, TI = 900 ms, TR = 2300 ms, and weighting = T1. The scan protocols were identical for all MRIs.

Subject characteristics

On the basis of the aforementioned criteria (2.2 MRI parameters), 94 samples were randomly obtained from the ADNI database. All subjects initially underwent several neuropsychological examinations resulting in several clinical characteristic indicators, including the MMSE score and the CDR. We randomly selected a total of 34 subjects [5 AD, 20 mild cognitive impairment (MCI), 9 healthy controls (HC)] (group 1) with an age range of 60–82 years (mean 74.01 ± 7.09). MMSE scores ranged from 20 to 30 (mean 25.94 ± 2.62) and the CDR scores ranged from 0 to 2 (mean 0.55 ± 0.39) (see Table 1) for IDM model. The 60 subjects (20 AD, 20 MCI, 9 HCs) (group 2) were employed to evaluate IDM model with ages ranged from 62 to 89 (mean 75.49 ± 5.29) years, while their MMSE and CDR scores ranged from 18 to 30 (mean 25.01 ± 2.32) and 0 to 2 (mean 0.58 ± 0.42), respectively (see Table 2). Data samples in groups 1 and 2 were age matched to exclude possible effects of brain atrophy related with age.

Methods

In this section, we describe the computational processes applied to the data. The pipeline of the proposed CAAD system is illustrated in Fig. 1. Firstly, we discuss the normalized gray matter ratios (NGMRs) obtained from the 3-Tesla 3D T1-weighted MRI data using the VBM approach. Secondly, IDM is applied to calculate the probability of AD based on percentage of GMV of samples in group 1. Thirdly, we validate the results of the proposed approach, and the correlations between AD risk values (IDM) and MMSE and CDR are estimated from subjects in group 2. Detailed explanations concerning the three steps are provided in the followingsections.

Global gray matter volume: VBM analysis

Data pre-processing was performed using statistical parametric mapping (SPM8, http://www.fil.ion.ucl.ac.uk/spm) and the voxel-based morphometry toolbox (VBM8, http://dbm.neuro.uni-jena.de/vbm) implemented in Matlab 2013a (MathWorks, Natick, MA). The VBM8 toolbox uses the unified segmentation model in which structural MRIs are bias-corrected and segmented into white matter (WM), GM, and cerebrospinal fluid (CSF) components [35]. More details of VBM analysis on sMRI is explained in [7]. In the current study, total GMV, WM volume (WMV), and CSF volume were obtained from the VBM8 toolbox based on unsmoothed segmented images, and total intracranial volume (TIV) was calculated as the sum of the GMV, WMV, and CSF volumes. In order to correct for variations in head size, the GMV of each subject was normalized by dividing the individual value by the TIV of the corresponding subjects. In this study, normalized gray matter datasets were used to generate AD risk assessment using IDM and evaluation.

IDM

In this study, the normal diffusion model was used for probability estimations [36]. Let Y ={ y ₁, y ₂, y ₃ ⋯ y _n } be a set of observations, called a given sample, and U ={ u ₁, u ₂, u ₃ ⋯ u _m } be the chosen framework space. If the observations cannot provide sufficient information to identify the precise relationship that is needed, then Y is called an incomplete data set. For any y ∈ Y and u ∈ U, Equation (1) is called normal information diffusion: $f_{j} (u_{i}) = \frac{1}{h \sqrt{2 π}} e^{- \frac{(y_{j} - u_{i})^{2}}{2 h^{2}}}$ (1)

Where h is the diffusion coefficient, which can be determined according to the maximum value b and minimum value a of the samples, and the sample number m in the sample set is as follows [36]: $h = 1.4208 \frac{(b - a)}{(m - 1)} m \geq 10$ (2)

Let $Di = \sum_{j = 1}^{m} f_{i} (u_{j})$ (3)

The distribution of sample information by Equation (4) was then generalized. $u_{x_{i}} (u_{j}) = \frac{f_{i} (u_{j})}{D_{i}}$ (4)

The function of u _xi (u _j) can be called the normalized information distribution of sample x _i. A good result for risk analysis can be obtained through treatment of the function u _xi (u _j). If we let x ₁, x ₂, … x _m be the m specified observation values, then the function can be called the information quantum diffused from the sample of X ={ x ₁, x ₂, … x _m } to the observation point of u _j. This can be represented as follows: $q (u_{j}) = \sum_{i = 1}^{m} u_{x_{i}} (u_{j})$ (5)

The physical meaning of the above function is that if the observation value of precipitation can only be chosen as one of the values in the series of u ₁, u ₂, … u _n, then the sample number with the observation value of u _j can be determined to be q (u _j) through the information diffusion from the observation set of x ₁, x ₂, … x _n, with regard to all values of x _j as the representatives of the samples. It is obvious that the value of q (u _j) is generally not a positive integer, but it is sure to be a number no less than zero. For the sum of q (u _j) values: $Q = \sum_{j = 1}^{m} q (u_{j})$ (6)

In fact, Q should be the summation of the sample number on each point of u _j. Therefore, the frequency of a sample falling at u _j can easily be estimated according to Equation (7). $p (u_{j}) = \frac{q (u_{j})}{Q}$ (7)

It is also obvious that the probability value transcending of u _j should be as follows: $p (u \geq u_{j}) = \sum_{j = 1}^{m} p (u_{j})$ (8)

Where p (u ≥ u _j) is the required risk estimate value. The GMVs obtained from samples in groups 1 were only used for IDM model.

Validation

The Pearson correlation test, using Statistical Package for Social Sciences software (SPSS version 20), was performed to check whether the result of risk assessment for AD was in line with the results of traditional methods, namely MMSE and CDR. MMSE and CDR are two common approaches in the diagnosis of AD based on surveys. In validation procedure, the GMVs obtained from group 2 wereemployed.

RESULTS AND DISCUSSION

Global volume of GM

TIV and GM obtained from VBM toolbox are expressed as mean ± standard deviation (SD) in volume (ml). In order to correct for variations in head size, the GMVs of all subjects were normalized by dividing the individual value by the TIV of the respective subjects. The NGMRs were used to generate AD risk assessment using IDM and evaluation. Table 3 shows TIV and gray matter volume obtained from group 1.

IDM

The probability of AD based on the NGMR using IDM in samples of group 1 is calculated and demonstrated in Fig. 2. According to the results, the risk of AD is boosted by a reduction of NGMR. For example, the risk of AD among people with 50% NGMR is about 5%. Similarly, the risk of AD for people with 45% NGMR is less than 30%. Patients with 40% NGMR are at risk of AD at a probability of 57%. The risk of AD among people with 32% NGMR is 90%.

According to the IDM results, the corresponding risk of AD based on percentage of NGMR is extracted and presented in Table 4. This table can be used as a guideline for the diagnosis of AD with regard to NGMR%.

Validation

The relationships between the results of the PCAADS and other methods (MMSE and CDR) were investigated using correlation analysis among group 2 subjects. The results of the means, standard deviations, and correlation matrices of the variables are presented in Table 5.

According to the results of the correlation test, the PCAADS had a significant and direct relationship with CDR (r = 0.659, p < 0.001). PCAADS has a significant and negative correlation with MMSE (r = –0.623, p < 0.001). CDR has a significant and negative correlation with MMSE (r = –0.785, p < 0.001).

These results indicated that the PCAADS results had a significant relationship, regardless of the sign, with MMSE and CDR. This means that the PCAADS conforms with the results of MMSE and CDR, two traditional survey-based methods.

The statistical results obtained by IDM showed that there is linear correlation in severity of GM volume atrophy and the risk of AD. In addition, correlation test between PCAADS and CDR or MMSE indicated that IDM may be a reliable technique for computerization risk assessment of AD especially with small data-sample sizes. In this research, the risk analysis for AD assessment was based on only NGMR, while AD is affected by various factors such as age, family history, apolipoprotein E (APOE)-4, genetics, body mass index, race, and ethnicity [1]. For example, African-Americans have higher rates in AD than whites. As an illustration, Japanese-American men have significant higher rates of AD, as compared to age-matched men in Japan. Hence, it is suggested as a future research direction for applying a fuzzy recognition model to estimate the risk of multifaceted phenomena [37].

Conclusion

This empirical study developed a mathematical approach with a computer-aided AD diagnosis system using IDM, which is an automated method for the detection of AD risk. The risk value of the subjects was determined by using the NGMR obtained from s-MRI data extracted by the VBM technique. According to the IDM results, the risk of AD increases with a reduction of GMV, which is illustrated as a linear relationship. The proposed model is a practical approach to the detection of patients with early AD. It is acknowledged that IDM is a useful theory for calculation of risk of natural phenomena (e.g., disease) with small data-sample sizes [32, 38]. This is important because the risk of AD differs based on ethnicity and it is less likely to be possible to collect large sample sizes for all races. Therefore, IDM contributes to the calculation of AD risk with smaller samples. The results of the proposed CAAD approach conform to the results of other common methods (CDR and MMSE). The main advantage of our model is that it is an automated approach that functions based on the volume of GM, while CDR and MMSE are focused on the self-reported results of surveys, which might be influenced by common method bias.

Disclosure Statement

Authors’ disclosures available online (http://j-alz.com/manuscript-disclosures/15-1176r1).

References

Alzheimer’s Association. What is Alzheimer’s? http://www.alz.org/alzheimers_disease_what_is_alzheimers.asp.

Matsuda

, Mizumura

, Nemoto

, Yamashita

, Imabayashi

, Sato

, Asada

(2012) Automatic voxel-based morphometry of structural MRI by SPM8 plus diffeomorphic anatomic registration through exponentiated lie algebra improves the diagnosis of probable Alzheimer disease. Am J Neuroradiol 33, 1109–1114.

Bron

, Smits

, van der Flier

, Vrenken

, Barkhof

, Scheltens

, Papma

, Steketee

RME

, Méndez Orellana

, Meijboom

, Pinto

, Meireles

, Garrett

, Bastos-Leite

, Abdulkadir

, Ronneberger

, Amoroso

, Bellotti

, Cárdenas-Peña

, Álvarez-Meza

, Dolph

, Iftekharuddin

, Eskildsen

, Coupé

, Fonov

, Franke

, Gaser

, Ledig

, Guerrero

, Tong

, Gray

, Moradi

, Tohka

, Routier

, Durrleman

, Sarica

, Di Fatta

, Sensi

, Chincarini

, Smith

, Stoyanov

, Sørensen

, Nielsen

, Tangaro

, Inglese

, Wachinger

, Reuter

, van Swieten

, Niessen

, Klein

(2015) Standardized evaluation of algorithms for computer-aided diagnosis of dementia based on structural MRI: The CADDementia challenge. Neuroimage 111, 562–579.

, Qin

, Gao

, Zhu

, He

(2014) Discriminative analysis of multivariate features from structural MRI and diffusion tensor images. Magn Reson Imaging 32, 1043–1051.

Liu

, Tosun

, Weiner

, Schuff

(2013) Locally linear embedding (LLE) for MRI based Alzheimer’s disease classification. Neuroimage 83, 148–157.

Cuingnet

, Gerardin

, Tessieras

, Auzias

, Lehéricy

, Habert

, Chupin

, Benali

, Colliot

(2011) Automatic classification of patients with Alzheimer’s disease from structural MRI: A comparison of ten methods using the ADNI database. Neuroimage 56, 766–781.

Beheshti

, Demirel

(2015) Probability distribution function-based classification of structural MRI for the detection of Alzheimer’s disease. ,. Comput Biol Med 64, 208–216.

Zhang

, Wang

, Phillips

, Dong

, Ji

(2015) Biomedical signal processing and control 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.

Beheshti

, Demirel

(2016) Feature-ranking-based Alzheimer’s disease classification from structural MRI. Magn Reson Imaging 34, 252–263.

10.

Beheshti

, Demirel

, Yang

(2015) Significance of sex differences on gray matter atrophy in Alzheimer’s disease: A voxel-based morphometry study. ,. Br Biomed Bull 3, 522–536.

11.

Andersen

, Rayens

, Liu

, Smith

(2012) Partial least squares for discrimination in fMRI data. Magn Reson Imaging 30, 446–452.

12.

Graña

, Termenon

, Savio

, Gonzalez-Pinto

, Echeveste

, Pérez

, Besga

(2011) Computer aided diagnosis system for Alzheimer disease using brain diffusion tensor imaging features selected by Pearson’s correlation. Neurosci Lett 502, 225–229.

13.

Lee

, Park

, Han

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

14.

Clerx

Lies

, Visser

Pieter Jelle

, Verhey

Frans

, Aalten

Pauline

(2012) New MRI markers for Alzheimer’s disease: A meta-analysis of diffusion tensor imaging and a comparison with medial temporal lobe measurements. J Alzheimers Dis 29, 405–429.

15.

Oishi

, Mielke

, Albert

, Lyketsos

, Mori

(2011) DTI analyses and clinical applications in Alzheimer’s disease. Adv Alzheimers Dis 2, 525–534.

16.

Vemuri

, Jack

(2010) Role of structural MRI in Alzheimer’s disease. Alzheimers Res Ther 2, 23.

17.

Patterson

, Feightner

, Garcia

, Hsiung

G-YR

, MacKnight

, Sadovnick

(2008) Diagnosis and treatment of dementia: 1. Risk assessment and primary prevention of Alzheimer disease. Can Med Assoc J 178, 548–556.

18.

Thompson

, Hayashi

, de Zubicaray

, Janke

, Rose

, Semple

, Herman

, Hong

, Dittmer

, Doddrell

, Toga

(2003) Dynamics of gray matter loss in Alzheimer’s disease. J Neurosci 23, 994–1005.

19.

Ashburner

, Friston

(2000) Voxel-based morphometry–the methods. Neuroimage 11, 805–821.

20.

Hirata

, Matsuda

, Nemoto

, Ohnishi

, Hirao

, Yamashita

, Asada

, Iwabuchi

, Samejima

(2005) Voxel-based morphometry to discriminate early Alzheimer’s disease from controls. Neurosci Lett 382, 269–274.

21.

Guo

, Wang

, Li

, Qi

, Jin

, Yao

, Chen

(2010) Voxel-based assessment of gray and white matter volumes in Alzheimer’s disease. Neurosci Lett 468, 146–150.

22.

špačková

, Novotná

, Šejnoha

(2013) Probabilistic models for tunnel construction risk assessment. Adv Eng Softw 62-63, 72–84.

23.

Chongfu

(2000) Demonstration of benefit of information distribution for probability estimation. Signal Processing 80, 1037–1048.

24.

Huang

(2002) An application of calculated fuzzy risk. Inf Sci (Ny) 142, 37–56.

25.

Shang

, Lu

, Jin

, Zhang

(2004) Unlimited information diffusion method and application in risk analysis in coronary heart disease. Int J Gen Syst 33, 233–242.

26.

Jiquan

, Xingpeng

, Zhijun

(2012) Natural disaster risk assessment using information diffusion and geographical information system. Handbook on decision making, Springer Heidelberg Berlin.

27.

Folstein

, Folstein

, McHugh

(1975) “Mini-mental state”. A practical method for grading the cognitive state of patients for the clinician. J Psychiatr Res 12, 189–198.

28.

Patil

, Ramakrishnan

(2014) Analysis of sub-anatomic diffusion tensor imaging indices in white matter regions of Alzheimer with MMSE score. Comput Methods Programs Biomed 117, 13–19.

29.

Morris

(1993) The Clinical Dementia Rating (CDR): Current version and scoring rules. Neurology 43, 2412–2414.

30.

Podsakoff

, MacKenzie

, Lee

J-Y

, Podsakoff

(2003) Common method biases in behavioral research: A critical review of the literature and recommended remedies. J Appl Psychol 88, 879–903.

31.

Huang

(2009) Integration degree of risk in terms of scene and application. Stoch Environ Res Risk Assess 23, 473–484.

32.

Hao

, Yang

, Gao

(2014) The application of information diffusion technique in probabilistic analysis to grassland biological disasters risk. Ecol Modell 272, 264–270.

33.

Mufson

, Mahady

, Waters

, Counts

, Perez

, DeKosky

, Ginsberg

, Ikonomovic

, Scheff

, Binder

(2015) Hippocampal plasticity during the progression of Alzheimer’s disease. Neuroscience 309, 51–67.

34.

Alzheimer’s Disease Neuroimaging Initiative, http://adni.loni.usc.edu/

35.

Cousijn

, Wiers

, Ridderinkhof

, Van den Brink

, Veltman

, Goudriaan

(2012) Grey matter alterations associated with cannabis use: Results of a VBM study in heavy cannabis users and healthy controls. Neuroimage 59, 3845–3851.

36.

Huang

(1997) Principle of information diffusion. Fuzzy Sets Syst 91, 69–90.

37.

, Guo

, Chen

, Zhou

(2006) Use of variable fuzzy sets methods for desertification evaluation. Comput Intell Theory Appl Int Conf 9th Fuzzy Days Dortmund, Ger Sept 18-20, 2006 Proc 721–731.

38.

Olya

HGT

, Alipour

(2015) Risk assessment of precipitation and the tourism climate index. Tour Manag {50, 73–80.