Transcript Alzheimer`s disease diagnosis based on SPECT brain

Eng.ª Biomédica
Instituto Superior Técnico, Faculdade de Medicina de Lisboa
Neuroimaging in
Alzheimer’s Disease
Diagnosis
Mafalda Gonçalves, nº 62857
Mafalda Negrão, nº 62847
Mariana Vieira, nº 59149
Tutora: Prof.ª Margarida Silveira
Marta Carvalho, nº 62838
2007/2008
Alzheimer’s disease
• Degenerative, progressive and fatal central nervous
system disease
• Most common form of dementia
• Difficulties communicating, learning and reasoning;
Anatomic level:
• Growth of the ventricles;
• Metabolic and blood flow
reductions in the
parietotemporal cortex.
Alzheimer’s diagnosis
An early diagnosis is important to improve patient’s life.
3 types of examination:
• Neuropsychological screening tests: Mini Mental State
Examination;
• Laboratory tests: Blood or urine;
• Imaging medical exams: MRI, CT and ET(SPECT and
PET).
Automated analysis helps the expert in the diagnosis
by clarifying information.
Purpose: study the efficiency of SPECT automation in
detecting AD.
Medical
Neuroimaging
in AD
MRI
MRI image is the result of the
magnetiztion of hydrogen atoms
CT
ET
Creates the image
by using a selection
of individual small
X-ray sensors and a computer.
Based on the
administration of a Tracer followed
by the measurement of radioactive
signal distributed in the brain.
PET
SPECT
A “gamma” camera detects
where the compound(signaled
by the tracer) has gone.
PET scanner detects the
gamma rays produced
by the collision of the
positrons with the electrons.
• By combining all of these methods it is posible to achieve a good
diagnosis.
Automated Analysis
•
Data Processing
Techniques
•
Classifiers
Allow the comparison of different
images since they transform them to
the same geometric space and
intensity.
A classifier uses a function to
discriminate the images available into
different classes.
1.
•
2.
3.
4.
5.
Voxel-Based SPECT and PET
analysis
Statistical Parametric Mapping
(SPM)
Three-Dimensional stereotactic
surface projection(3D-SSP)
Tomography z-Score Mapping
Partial-Volume Correction
(PVC)
•
•
Linear and supervised:
- an optimal linear classifier;
Nearest Mean Classifier.
Linear and nonsupervised:
- Support Vector Machine.
Nonlinear:
- Artificial neural networks
C
Classifiers
An Optimal Linear Classifier
Nearest Mean Classifier
• linear and supervised
method
• calculates the means of the
feature vectors in the training
set for the different classes.
• Variable 1: average of voxels A, B, C, D and E
minus the average of voxels F, G, H and I.
• Variable 2: ratio of cerebellum to the average of
voxels A, B, C, D and E.
• Variable 3: average of voxels A, B, C, D and E.
• equidistant hyperplane
between the different classes
means
Support Vector Machine (SVM)
• plane midway between bounding planes
to separate the different classes.
• little or no spatial information about the
imaging problem
The Continuous Linear SVM
uses an adjacency matrix to represent the
relation among the features according to a
relation function.
Artificial Neural Networks
The OINN (the optimal interpolative
neural network) is trained and tested by
vectors composed of a patient’s regions
of interest (ROIs).
Conclusion
Scientists have been studying different techniques to diagnose AD,
developing methods that allow to discriminate Alzheimer’s Dementia
controlled by an automated analyses of SPECT Volumes.
Classifiers can provide a quantitatively automatic analysis of the
SPECT volumes.
• good choice of features (characterizes the different classes)
• discrimination between classes
an independent and a better diagnosis