1. REPLICATION AND IDENTIFICATION OF
CLINICAL SUBGROUPS IN A STRUCTURAL
MRI DATASET FOR SCHIZOPHRENIA
Dakarai McCoy (PM), Abeer Jihad, Edith Castellanos
Sponsors: Drs. Vince Calhoun (MRN), Jessica Turner (GSU)
Technical Mentor: Navin Cota

2. Objective
To replicate the results obtain by the Mind Research Network
of clinical subtyping in Schizophrenia by finding the
relationship between the Gray Matter concentration (GMC)
and clinical scores.

3. Purpose
The success of this research will discover reliable clinical subtype patterns
of Gray Matter Concentration (GMC) deficit, which could be targeted in
the personalized development of drugs for schizophrenia.
• Schizophrenia is a chronic brain disorder
• Delusion
• Hallucination
• Disorganized speech
• Catatonic behavior
• Negative symptoms

4. Structural Magnetic Resonance Imaging sMRI
• Body is composed of 70% H2O
• Spinning charged particle
• Magnetic moment

5. Hydrogen Atom
• Randomly oriented with no applied field
MRI components
• Primary magnet
• Gradient magnets
• Radiofrequency (RF)coils
• Computer System

6. Primary Magnetic Field
• Hydrogen atoms align parallel or antiparallel
to the primary field (B0)
• Precession

7. Gradient Coils
• Allow spatial encoding for MRI images in the x
y, and z axis
• RF pules
• Signal received by the RF coil in the transverse
plane

8. Computer System
White Matter (WM)light gray Cerebrospinal fluid (CSF) black
Gray Matter (GM) dark gray

9. Structural Magnetic Resonance Imaging sMRI
T1-Weighted

10. Source Based Morphometry
T1
VBM ICA
GMC
WM
CSF
Source Based Morphometry

11. Independent Component Analysis ICA
Maximally independent sources (two groups; Controls and Sz)
Column of loadings for a given source map(say Insula) covariation
among subjects is captured in loadings.
For a given component map (say insula) if there are two groups, if
mean of Controls in A1 is greater than mean of Sz in A1; then it
means Ct has more GMC than Sz
Top Two Components (HC>Sz) from
C.N.Gupta et al based on Effect Size,
Schizophrenia Bulletin, 2014.
Voxel Component
Insula,STG
Voxel
Subject
Subject
A1
Component
*
A2
C1
C2
Frontal
Component
Matrix
Ctr and Szs GMC
matrix regressed
of Age, Gender, site
voxelwise
Loading
Matrix

12. • Develop MATLAB code to extract Sz patients
from a data set.
• Data extracted: URSI, site, age, sex, SAPS/SANS.
Module 1
Extraction of patients
with Schizophrenia
• Perform multivariable regression of age and
gender to make the results more sensitive to
group differences (Cota et Al 2009)
Module 2
Regression of
covariates
•. Run ICA using SBM toolbox from GIFT
•Perform B-ICA in the selected components
•Perform 2 sample t-test
Module 3
ICA and Biclustering
Method

13. Module 1 Extraction of patients with Schizophrenia
• 109 Schizophrenia patients from 1st data set
• Variables extracted: Site, Sex, Age, SAPS/SANS scores

14. Module 2 Regression of covariates
• Extracted brain images from 109 patients
• Run regression of covariates [Age, Sex, Site]
• Smooth selected images
Picture of the brain of one subject after regression
Picture of the brain of one subject after smoothing

15. Module 3 ICA • Used SBM to run ICA for 109 images
• 30 components. Components selected 5 and 7
Medial frontal gyrus (mFG)
Superior frontal gyrus (SFG)
Middle fontal gyrus (MFG)
Insula (I)
Inferior frontal gyrus (IFG)
Superior temporal gyrus (STG)
Component 5
Component 7
Sagittal (y) Frontal/coronal (x) Transverse/Axial (z)

16. Voxel Component
Insula,STG
Voxel
Subject
Subject
A1
Component
*
A2
C1
C2
Frontal
2. INTERSECTION OF ABS SORTED LOADINGS
A1
A2
3. SUBGROUPS DECIPHERED
AND A TWO SAMPLE TTEST IS
PERFORMED ON THE
CLINICAL SYMPTOMS
109 Szs GMC matrix regressed
of Age, Gender, site voxelwise
with two overlapping biclusters
S1
S2
Sinter
Sorted by Abs value. Gradient shows the
higher magnitude (negative and positive)
loadings pushed up
Rule of thumb: Dissect Subject
number by 4 and do an intersection
of sorted subject names in A1 and
A2
Loading
Matrix
Component
Matrix
Biclustering (B-ICA)

17. Preliminary Results
• 30 subjects from top quarter
• S1 = 16, S2 = 16 , S(inter) = 11 , S(total) = 27
• Scale for the Assessment of Positive Symptoms (SAPS)
• H = 0, p < 0.6875
• Scale for the Assessment of Negative Symptoms (SANS)
• H = 0, p < 0.736

18. Future Work
• Increase number of subjects – (4 more data sets)
• Integration into Genetics project

19. Questions

20. Thank you!