Conference PaperPDF Available

Unveil the hidden Information behind the variables of Alzheimer's disease (AD): a systematic comparison of Manifold Learning algorithms in AD

Authors:
Unveil the Hidden Information behind the Variables of Alzheimer's Disease (AD): a
Systematic Comparison of Manifold Learning Algorithms in AD
1Department of Computer Science, The University of Western Ontario, London, ON, Canada; 2Robarts Research, London, ON, Canada; 3Lawson Health
Research Institute, St. Josephs Healthcare, London, ON, Canada; 4Division of Geriatric Medicine, University of Western Ontario, London, ON, Canada
Peng Dai1,2, Femida Gwadry-Sridhar1,3, Michael Bauer1, Michael Borrie4
References:
[1] P. Dai, et al., Structural Differences in Cognitively Normal, Mild Cognitive Impairment, and Alzheimer's Disease Individuals: A Novel
Study Based on Brain Symmetry, in AAIC, Washington D.C., USA, July 2015.
[2] P. Dai, et al., A hybrid manifold learning algorithm for the diagnosis and prognostication of Alzheimer's disease, in AMIA 2015 Annual
Symposium, San Francisco, CA, USA, Nov 2015.
0 5 10 15 20 25 30 35 40 45 50
-3
-2
-1
0
1
2
3
4
20 40 60 80 100 120 140
20
40
60
80
100
120
140
-30 -20 -10 0 10 20 30
-15
-10
-5
0
5
10
15
Introduction
Alzheimer's disease (AD) is a chronic neurodegenerative disease causing dementia, amnesia and deficit in one or more cognitive
functions, which affects an individual’s ability to carry out activities of daily living (ADLs). Statistical estimation of whether an individual
has AD involves the analysis of various physiological variables, e.g. images (Magnetic Resonance Imaging (MRI), Positron Emission
Tomography (PET)), genomics, metabolism, etc. Statistical AD diagnosis can be formulated as a multiple class classification problem
in machine learning. Much of the research in this area makes direct use of the raw values of variables in the statistical analyses,
which are usually contaminated with noise and distortion. A manifold learning step can be introduced to remove noise and extract
discriminant features (or variables) for the statistical analyses of AD.
Objective
Despite the fact that various machine learning algorithms have been investigated for the automatic diagnosis of AD, limited attention
has been put on the design of an optimal manifold, i.e., one which has the best discriminant ability. This study explores this property
of various manifold learning algorithms in an automatic diagnosis framework. In particular, we focus on the evaluation of different
manifold learning algorithms in terms of AD diagnosis.
Methods
Evaluation tests are carried out using the neuroimaging and biological data from the Alzheimer's Disease Neuroimaging Initiative
(ADNI) in a three-class (normal, mild cognitive impairment, and AD) classification task using support vector machines (SVM). The
imaging and biological data from 843 patients from ADNI are adopted for verification tests. The MRI images are registered to a unified
brain model and transformed to stereotaxic space, followed by tissue classification and brain volume calculation using CIVET [1][2].
Five different manifold learning algorithms were chosen for comparison: Locality Preserving Projection (LPP), Principal Component
Analysis (PCA), Neighborhood Preserving Embedding (NPE), Stochastic Proximity Embedding (SPE) and Sammon mapping.
Tenfold cross validation is utilized in our experiment setup.
Acknowledgements:
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.
Results
In our randomized verification tests, manifold learning algorithms clearly improve the performance of
the automatic diagnosis task. Without manifold learning, the SVM based automatic diagnosis system
obtains an average diagnosis accuracy of 76.67% (optimal at 30 selected features), while all the
manifold learning algorithms outperform the baseline by 2% to 17%. In particular, the Neighborhood
Preserving Embedding (NPE) shows the best result, with 94.01%, accuracy with only 18 features.
Moreover, the optimal results are all from a subset of the entire variable set.
NPE shows the best performance with 18
selected features.
Eigen decomposition on
Conclusion
Manifold learning is an effective way to remove noise and extract discriminant features for
classification tasks, i.e. AD diagnosis. This can be a meaningful way to improve the performance of
automatic diagnosis systems. Considerations should be given to which algorithm is more suitable for
AD diagnosis. In addition, the experimental results show that there are strong correlations between
different variables utilized for AD diagnosis. Even a naïve dimension reduction approach can show
some improvements.
Fig. 1: Embedding multivariate medical records into a manifold [1][2].
Fig. 3: Experimental results for different manifold learning algorithms.



    
   
Fig. 2: System Diagram.
ResearchGate has not been able to resolve any citations for this publication.
ResearchGate has not been able to resolve any references for this publication.