Early Detection of Mild Cognitive Impairment Using Nonlinear ...

jamaicacooperativeΤεχνίτη Νοημοσύνη και Ρομποτική

17 Οκτ 2013 (πριν από 3 χρόνια και 19 μέρες)

98 εμφανίσεις



Early Detection of Mild Cognitive Impairment Using Nonlinear Analysis of EEG via
Tsallis Entropy and Support Vector Machine Classification

Munro, N. B.
, T. J. De Bock
, S. Das
, M. Mohsin
, J. McBride
, X. Zhao
, L. M. Hively
, C
, G. Jicha
L. Broster
and Y. Jiang

Oak Ridge National Laboratory;
University of Tennessee
Knoxville Department of
Mechanical, Aerospace, and Biomedical Engineering;
University of Kentucky College of




An urgent research priority involves efficient and accurate means to identify preclinical

Alzheimer’s disease (AD) in older adults with memory complaints. The current gold
standard methods (e.g., MRI, PET) are expensive and not routinely used.
A substantial
increase in dementia cases is expected with unsustainable costs unless early detection
becomes possible and is paired with intervention to slow or halt cognitive decline. Oak
Ridge National Laboratory (ORNL) has partnered with the University

of Kentucky (College of
Medicine) and the University of Tennessee on a pilot project for early detection of mild
cognitive impairment (MCI) via nonlinear analysis of EEG. The data were acquired at UK
during a simple non
task protocol, followed by a hybrid

working memory (WM) task. We
Tsallis entropy
based quantified EEG (qEEG) analysis of

task data
. The results
were encouraging (100% sensitivity, 89% specificity, 94% accuracy), although the sample
size was small (N = 9 normal, 7 MCI). We also hav
e obtained additional encouraging
results in initial Support Vector Machine analyses of WM data (N = 17 normal, 17 MCI).
This analysis used Tsallis entropy and other conventional statistical measures as features,
yielding 82% accuracy, 88% sensitivity and

76% specificity. Further work is needed to
increase sample size, improve discrimination, minimize the number of electrodes, and
eliminate the WM task for ease of use. The objective is implementation of the best
algorithm(s) on a desktop or laptop compute
r to give real
time analysis results for use in
the primary care or community hospital setting.

Biographical Sketch:

A native of Columbus, Ohio, Dr. Munro is a graduate of Capital University with a B.S. in
Biology. Dr. Munro is a neuroendocrinology
rained physiologist whose Ph.D. was earned in
the Department of Physiology and Biophysics within the University of Kentucky College of
Medicine. She was the PI on the ORNL Seed (Laboratory Directed Research and
Development) project under which our pilot E
EG data and qEEG results

were obtained. She
has many years’ experience coordinating multidisciplinary projects at ORNL, doing project
management, and writing extensive review papers and technical reports. Much of her work
centered on evaluating the potent
ial health effects of chemicals and radionuclides and on
the toxicology of chemical warfare agents, their antidotes, and degradation products. She
has collaborated for over a decade on research into the forewarning of condition change in
various biomedica
l data such as EEG for forewarning of epileptic seizures. She is retired
from ORNL and working on a consulting basis with researchers at UK, ORNL,

and UT