Language Understanding Analysis in Schizophrenic Patients from MEG Data Keshab K. Parhi, ECE Tingting Xu, ECE Massoud Stephane, Psychiatry, VA Medical Center Arthur Leuthold, Neuroscience, VA Medical Center Feb. 16, 2012
Welcome message from author
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Language Understanding
Analysis in Schizophrenic
Patients from MEG Data
Keshab K. Parhi, ECE
Tingting Xu, ECE
Massoud Stephane, Psychiatry, VA Medical Center
Arthur Leuthold, Neuroscience, VA Medical Center
Feb. 16, 2012
Outline
Background & Objective
Design & Methodology
Results & Analysis
Conclusion & Future Work
Schizophrenia
Chronic, severe, disabling mental disorder
About 1% Americans affected
Causes remain unknown
Symptoms
Positive Symptoms
Hallucination
Delusion
Negative Symptoms
Disruptions to normal emotions and behaviors
Lack of pleasure in everyday life
Cognitive Symptoms
Problems with attention, memory, decision making
Language disorder
Objective
Understand abnormal neural oscillations (frequency, location, time) associated with language disorder in schizophrenia.
Level-specific neural oscillations
Neural dynamics across language levels
Linguistic Task
Discriminate correct & incorrect language stimuli
Condition
Stimulus
Correct Incorrect
Lexical The-cabin-fire-roped-the The-cabin-freet-roped-the
Semantic The-boy-ate-the-bagel The-bagel-ate-the-boy
Discourse
My sister wanted to replace a kitchen cabinet. She called a local
company and gave them the desired dimensions…
The-company-built-the-cabinet The-company-bought-the-kitchen
Classification based feature selection
Level-specific (same condition)
Neural-dynamic (across conditions)
Outline
Background & Objective
Design & Methodology
Results & Analysis
Conclusion & Future Work
Feature Extraction (ERD/ERS)
BPF down-sample
MEG calculate power
1- 48 Hz 8 bands
smooth feature
256 KHz
calculate ERD/S
250ms window 125ms overlap
mean power
𝐸𝑅𝐷𝑆(𝑗) = 𝐴(𝑗) − 𝑅
𝑅∗ 100% 𝑅 =
1
𝐾 𝐴(𝑗)
𝑛0+𝐾
𝑗=𝑛0
Spectral-Temporal-Spatial Feature Set 248 channel * 8 freq. band * 92 time points = 182,528 features
Linguistic Task
Timing diagram (11.5 sec / trial)
3 sec 5 sec 3.5 sec
Baseline Encode Post-stimuli
1 2 3 4 5
Stimulus (Word / Nonword)
Correct: the - cabin - fire - roped – the
Incorrect: the - cabin - freet - roped - the
Framework
Language
Task
Record
MEG
raw
MEG BPF
subband
MEG Calculate
ERD/S
Fscore
Filtering
full
feature set
SVM-RFE
Backward
Selectiong
top
features
Feature Selection
Analysis
Classification
Subjects
12 schizophrenia patients, 10 healthy controls
Native English speaker, right-handed
Patients meet DSM-IV criteria
Experimental protocol was approved by the Institutional Review Boards of the VA Medical Center and of the University of Minnesota
ERD/ERS – Delta 1-4Hz
ERD/ERS – Alpha 8-12Hz
ERD/ERS – Gamma 40-48Hz
Outline
Background & Objective
Design & Methodology
Results & Analysis
Conclusion & Future Work
Level-specific result
0 10 20 30 40 50 60 70 80 90 1000
10
20
30
40
50
60
70
80
Number of Features
Pre
dic
tion E
rror
(%)
Lexical
Semantic
Classification error
Classification (LOO cross validation )
Lexical: 10 Controls, 12 Patients , 20 top features
Semantic: 11 Controls, 10 Patients, 53 top features
Top Feature Analysis (lexical)
Delta band, occipital lobe, post-stimuli phase
Top Feature Analysis (lexical)
Alpha band, left-temporal lobe, encoding phase
Top Feature Analysis (lexical)
Beta band, left-frontal lobe, encoding phase
Across-level result
Control: sem (10) v.s. dis (9), 20 top features
- LOO cross validation error 3/19
- top features cover multiple bands and at different brain locations
Patient: sem (10) v.s. dis (7), 5 top features
- LOO cross validation error 1/17
- most top features are selected from 12-16Hz frequency band and located at left temporal lobe.
Top features (patient)
Top features (control)
Conclusion
Evaluate abnormal neural oscillation in schizophrenia during language processing
Spectral-spatial-temporal MEG patterns
Two-step feature selection
High classification accuracy
Significant between group oscillation difference
Further Work
Neural dynamics between language levels
Larger dataset
More reliable feature selection algorithm
Acknowledgement
Ms. Tingting Xu, ECE
Dr. Massoud Stephane Psychiatry & Psychology , University of Minnesota
Brain Science Center, VA Medical Center
Dr. Arthur Leuthold Neuroscience, University of Minnesota
Brain Science Center, VA Medical Center
- MEG data collection
Minnesota Super Computing Institute (computation power)
Questions?
Related Documents
