![](bg1.png)
3D Local Binary Pattern for PET Image Classification by SVM
Application to Early Alzheimer Disease Diagnosis
Christophe Montagne, Andreas Kodewitz, Vincent Vigneron, Virgile Giraud and Sylvie Lelandais
University of Evry, IBISC Laboratory, 40 Rue du Pelvoux, CE 1455, 91020 Evry Cedex, France
Keywords:
Local Binary Pattern, Feature Extraction, Positron Emission Tomographic images, Alzheimer disease,
Machine Learning.
Abstract:
The early diagnostic of Alzheimer disease by non-invasive technique becomes a priority to improve the life
of patient and his social environment by an adapted medical follow-up. This is a necessity facing the growing
number of affected persons and the cost to our society caused by dementia. Computer based analysis of
Fluorodeoxyglucose PET scans might become a possibility to make early diagnosis more efficient. Temporal
and parietal lobes are the main location of medical findings. We have clues that in PET images these lobes
contain more information about Alzheimer’s disease. We used a texture operator, the Local Binary Pattern,
to include prior information about the localization of changes in the human brain. We use a Support Vector
machine (SVM) to classify Alzheimer’s disease versus normal control group and to get better classification
rates focusing on parietal and temporal lobes.
1 INTRODUCTION
The number of people affected by the Alzheimer Dis-
ease (AD) is growing. In 2010, this dementia af-
fected 35.6 million people (Wimo and Prince, 2010).
This disease affects patient himself and his social
environment. The detection of the early states of
Alzheimer disease allows to begin some treatments
for the patient and slow down the progress of the dis-
order (Salmon, 2008). Most of works focused on tem-
poral lobes, few of them on parietal lobes (Kodewitz
et al., 2011). Some works on the SPECT-scans (Sin-
gle Photon Emission Computed Tomography) are due
to Ramìrez with a computer-aided diagnosis based
on a selection of image parameters (first and second
order statistics) and Support Vector Machine (SVM)
(Ramìrez et al., 2009). Others methods with PET-
scans (Positron Emission Tomography) are based on
covariance analysis of voxels to classify AD versus
Normal Control (NC) (Scarmeas et al., 2004). The
main idea of this work is that we can automatically
analyse the PET scans to detect AD patient versus
NC patient using textural features. This classifica-
tion can help doctors to make their early diagnostic
and try to localize some brain areas where Alzheimer
Disease is growing. In our study, we use 3D-scans
extracted from the ADNI database (Alzheimer’s Dis-
ease Neuroimaging Initiative) and we propose to use
a textural operator which is the Local Binary Pattern
(LBP) (Pietikäinen and Ojala, 2000) extended to the
3D. Then this new feature was studied by an ANaly-
sis Of VAriance (ANOVA). The analysis highlighted
a link between LBP patterns and the type of patient
or some brain areas. Based on this new feature, we
train a machine learning algorithm SVM to classify
AD versus NC PET-scans. For this step, in a first time
we use as Region Of Interest (ROI) parietal and tem-
poral lobes, and then, in a second time full PET-scans
are considered.
In the first section we describe the materials used
in this study. Secondly we introduce the 3D Local
Binary Pattern that we developed. Then we explain
the ANOVA method and the obtained results. Finally
after a short presentation of SVM, we discuss on the
results of the classification using the LBP caracteris-
tics and we conclude.
2 MATERIALS
In this study, PET-scan provides three-dimensional
functional imaging data that measures the metabolism
in the brain which are acquired by a non-invasive
method, Figure 1 shows how the AD reduces the
metabolism in the brain. This scan is obtained by 18F-
145
Montagne C., Kodewitz A., Vigneron V., Giraud V. and Lelandais S..
3D Local Binary Pattern for PET Image Classification by SVM - Application to Early Alzheimer Disease Diagnosis.
DOI: 10.5220/0004226201450150
In Proceedings of the International Conference on Bio-inspired Systems and Signal Processing (BIOSIGNALS-2013), pages 145-150
ISBN: 978-989-8565-36-5
Copyright
c
2013 SCITEPRESS (Science and Technology Publications, Lda.)