Nancy B. Munro, ORNL, retired

1
Healthy Heart, Healthy Brain Early Alzheimers
Detection and Prevention
Nancy B. Munro, ORNL, retired Lee M.
Hively Computational Sciences and Information
Division Yang Jiang, University of Kentucky
College of Medicine Charles D. Smith, MD, UK
College of Medicine Gregory A. Jicha, MD, UK
College of Medicine Xiaopeng Zhao, University of
Tennessee Oak Ridge, Tennessee March19, 2013
2
Acknowledgements
University of Kentucky David Wekstein, PhD,
William Markesbery, MD (dec.) Adam Lawson and
several other doctoral students Juan Li, PhD, UK
Chinese Academy of Sciences, Institute of
Psychology, Beijing, China Luke Broster, MD/PhD
student University of Tennessee Joseph
McBride, PhD student Thibaut de Bock, Satyajit
Das, Maruf Mohsin, BME students (2009-10 senior
design project) Robert Sneddon, PhD W. Rodman
Shankle, MD
3
Outline

• Alzheimers disease
• Alzheimers early detection
• Gold standard methods
• EEG analysis approach, results
• Vision, future work
• Prevention through delay

4
Rationale

• Alzheimers
• 5.4 million Americans today
• 16 million expected by 2050
• Current costs
• 200 billion in 2012 1.5 trillion estimated
  by 2050
• 70 cared for at home
• Diagnosis of exclusion confirmation only at
  autopsy
• Value of early diagnosis
• - Early intervention
• - Tool for drug discovery

5
Alzheimers Disease

• Alzheimers late onset
• - onset age 60 and up
• - 4-20 year course to death
• Alzheimers early onset
• - onset in 40s, 50s
• - 4-8 year course to death
• Mixed
• - Alzheimers and vascular dementia
• - Alzheimers and DLB

6
New Drugs Hypotheses (Summers Therapy Sept.
2011)

• Amyloid
• Tau protein
• Inflammation
• Oxidative Stress
• Vascular
• Disordered glucose metabolism

7
New Drugs

• Amyloid-blocking (immunization other) failure
  to improve late-stage AD in trials
• Insulin via injection or nasal powder inhalation
  improved memory
• FDA requirement efficacy for function and
  cognition

8
Diagnosis via Analysis of Scalp EEG

• Current approaches costly, invasive
• MRI
• PET
• Neuropsychological testing
• Spinal tap for biomarkers amyloid, tau
• EEG
• Non-invasive
• Simple
• Inexpensive
• Rapid results

9
Experimental Design

• Groups Normal, early MCI, early AD
• Goal for N 20/group
• Protocols Min
• ORNL simple 30
• Working memory 15
• Total 45

10
Why Working Memory Task?
Changes take place earliest in brain areas of
short-term memory and progress

11

12
Actual Numbers Acquired and Analyzed

• Groups Simple WM
• Normal 21 (15) 17
• Early MCI 21 (16) 18
• Early AD 18 (17) 11

13
Intra-Individual Variability

• Minimize by
• All EEGs at same time of day
• All subjects at ease
• Same mental activity during protocol
• No ApoE4 allele
• No co-existing brain conditions
• No psychoactive drugs
• Well-matched age (76)
• education (17yr)

14
Simple ORNL EEG Protocol
Attach electrodes in standard 19-channel
montage, then record scalp EEG - 5 minutes
eyes open - 10 minutes eyes closed, counting
silently backwards while tap finger on each
count - 10 minutes eyes closed, awake - 5
minutes eyes open - 30 minutes total
De-identify, convert data to ASCII format UK
Data quality check ORNL
15
Results UT, Resting EEG Data
Method MCI vs. NC, (significance) AD vs. NC, (significance) MCI vs. AD, (significance)
Dynamic spectral and entropy features 86.7 (p lt 0.004) 93.3 (p lt 0.0001 90.0 (p lt 0.0006)
Multi-scale sample entropy (MSE) 93.5 (p lt 0.0001) 96.9 (p lt 0.0001 90.9 (p lt 0.005)
MSE (memory task) 90.9 (p lt 0.0001) 96.3 (p lt 0.0001) 100 (p lt 10-5)
Graphical EEG coherence analysis 93.6 (p lt 0.0003) 93.8 (p lt 0.0003) 97.0 (p lt 0.0003)
16
Conclusions

• Can discriminate normal from MCI and AD
• Both via ERP and advanced EEG analyses
• Nonlinear analysis both of WM and resting EEG
  data show promise
• Work ongoing on resting EEG data
• Further work needed for clinical utility

17
Future Work

• Acquire data from more participants
• Continue to improve analyses UT
• Apply ORNL graph-theoretic method
• Enhance accuracy with few electrodes
• Implement on laptop or PDA

18
Vision

• A device usable in
• Primary care setting
• Community hospitals
• For drug discovery
• Adapt for other neurodegenerative diseases
• Diffuse Lewy Body Disease
• Parkinsons Disease
• Fronto-temporal dementia

19
Prevention

• Risk factors Not Controllable
• Age
• Family history
• Genetic makeup
• Risk factors Controllable
• Smoking
• High blood pressure
• High cholesterol
• Poorly-controlled diabetes
• Depression
• Sleep apnea
• Lack of exercise
• Poor diet/obesity
• Education/cognitive inactivity

20
Prevention

• Eliminate preventable risk factors, e.g., smoking
• Exercise
• New neurons in hippocampus (memory area)
• Regular exercise reduces AD incidence
• Cognitive activity new neuronal connections
• Study foreign language
• Learn to play musical instrument
• Brain games (crosswords, Sudoku, etc.)
• Mindfulness meditation
• Diet rich in antioxidants, not pills
  Mediterranean (combats inflammation)

21
Healthy Aging
Good physical health Great aging brain ?
Regular physical exercise ? Positive emotions ?
Positive relationships ? Limiting chronic stress
Memory and the Aging Brain. Steven W. Anderson,
PhDThomas J. Grabowski, Jr. MD The University of
Iowa. June 2003
22
Prevention Summary

• Prevention through delay
• Whats good for your heart and lungs is good for
  your brain!

23
(No Transcript)
24
Questions?
25
Backup Slides
26
Hybrid Working Memory Task

Subjects were asked to hold the sample target
object in mind and indicate whether each test
object was the same as or different from the
sample object by pressing one of two buttons
using their Right or Left hand.
27
Results UK
Event-Related Potential (ERP) Analysis

The MCI group is similar in accuracy of memory
(above) to normal (NC), but ERPs (on right) of
MCIs were identical to those of ADs (blue arrows,
L frontal).
28
Sensitivity and Specificity

• Sensitivity ability to identify positive
  results TP
• TP FN
• Specificity ability to identify negative
  results TN
• TN FP

29
Results UT, WM Task Data
Support Vector Machine (SVM) Analysis

• Features 12 Tsallis entropies for each brain
  region
• Radial basis kernel function
• Accuracy 82
• Sensitivity 88
• Specificity 76

SVM analysis. An example of SVM classification
using a radial basis kernel function. The
features are averaged Tsallis entropy values of
the frontal sites (abscissa) and that of left
temporal sites (ordinate) N 0, MCI 1.
30
ORNL Advanced Analysis

• Graph-theoretic analysis under development
• Uses existing ORNL technology to filter data and
  construct phase-space diagram
• From that, network (graph) constructed and
  analysis performed
• Performs extremely well for seizure FW
• Must be adapted to group comparisons