a plan for brain connectivity analysis john melonakos
DESCRIPTION
A Plan for Brain Connectivity Analysis John Melonakos. Schizophrenia. Kandel, Schwartz, Jessell. “Principles of Neural Science, 4 th Edition.” (2000). p.1188. The Plan. Segment brain into white matter, gray matter, and CSF - PowerPoint PPT PresentationTRANSCRIPT
![Page 1: A Plan for Brain Connectivity Analysis John Melonakos](https://reader035.vdocuments.mx/reader035/viewer/2022081519/56813ad2550346895da30039/html5/thumbnails/1.jpg)
1National Alliance for Medical Image Computing http://na-mic.org
A Plan for Brain Connectivity Analysis
John Melonakos
![Page 2: A Plan for Brain Connectivity Analysis John Melonakos](https://reader035.vdocuments.mx/reader035/viewer/2022081519/56813ad2550346895da30039/html5/thumbnails/2.jpg)
2National Alliance for Medical Image Computing http://na-mic.org
Schizophrenia
Kandel, Schwartz, Jessell. “Principles of Neural Science, 4th Edition.” (2000). p.1188
![Page 3: A Plan for Brain Connectivity Analysis John Melonakos](https://reader035.vdocuments.mx/reader035/viewer/2022081519/56813ad2550346895da30039/html5/thumbnails/3.jpg)
3National Alliance for Medical Image Computing http://na-mic.org
The Plan
1) Segment brain into white matter, gray matter, and CSF
2) Divide resulting gray matter segmentation into key anatomical regions (e.g. the DLPFC)
3) Grow DTI fibers from the key anatomical regions to analyze connectivity
![Page 4: A Plan for Brain Connectivity Analysis John Melonakos](https://reader035.vdocuments.mx/reader035/viewer/2022081519/56813ad2550346895da30039/html5/thumbnails/4.jpg)
4National Alliance for Medical Image Computing http://na-mic.org
STEP 1: Find WM,GM,CSF
To do this we have chose an approach based on Bayesian Segmentation
Step 1
Data: Probabilities generated by applying a distribution (typically Gaussian) to your dataPriors: An initial guess at the solution
Posteriors: The resulting probabilities
Constant
Priors*DataPosteriors
![Page 5: A Plan for Brain Connectivity Analysis John Melonakos](https://reader035.vdocuments.mx/reader035/viewer/2022081519/56813ad2550346895da30039/html5/thumbnails/5.jpg)
5National Alliance for Medical Image Computing http://na-mic.org
The Power of Bayes’ Rule
Mumford, “The Bayesian Rationale for Energy Functionals”
I))|log(p(- )E(
Step 1
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10
1
2
3
4
5
6
7
8
9
10
E(w
)
p(w|I)
Minimizing the Energy = Increasing the Posterior Probability
scene thedescribe toused variables theare
![Page 6: A Plan for Brain Connectivity Analysis John Melonakos](https://reader035.vdocuments.mx/reader035/viewer/2022081519/56813ad2550346895da30039/html5/thumbnails/6.jpg)
6National Alliance for Medical Image Computing http://na-mic.org
Bayesian/Energy Relation
)(E)I,(E
))log(p(-))|-log(p(I
I))|log(p(- )E(
pd
Step 1
![Page 7: A Plan for Brain Connectivity Analysis John Melonakos](https://reader035.vdocuments.mx/reader035/viewer/2022081519/56813ad2550346895da30039/html5/thumbnails/7.jpg)
7National Alliance for Medical Image Computing http://na-mic.org
The Algorithm
• Goal: Segment Volume into 3 classes• Solution:
1. Create 3 Data terms
2. Guess at 3 Prior terms
3. Apply Bayes’ Rule 3 times
4. Find the maximum of the 3 resulting posteriors to determine the winning class
5. Apply a label for the winning class
Haker, et al. “Knowledge-Based Segmentation of SAR Data with Learned Priors” (1999)Teo, et al. “Creating connected representations of cortical gray matter for functional MRI visualization” (1998)Teo, et al. “Anisotropic diffusion of posterior probabilities” (1997)
Step 1
![Page 8: A Plan for Brain Connectivity Analysis John Melonakos](https://reader035.vdocuments.mx/reader035/viewer/2022081519/56813ad2550346895da30039/html5/thumbnails/8.jpg)
8National Alliance for Medical Image Computing http://na-mic.org
Added Tricks
• Goal: Segment Volume into ‘N’ classes
• Solution: 1. Create ‘N’ Data terms
2. Guess at ‘N’ Prior terms
3. Apply Bayes’ Rule ‘N’ times
4. Find the maximum of the ‘N’ resulting posteriors to determine the winning class
5. Apply a label for the winning class
Smooth posteriors
before finding the maximum
Iterate multiple times to refine the data and prior terms
Step 1
![Page 9: A Plan for Brain Connectivity Analysis John Melonakos](https://reader035.vdocuments.mx/reader035/viewer/2022081519/56813ad2550346895da30039/html5/thumbnails/9.jpg)
9National Alliance for Medical Image Computing http://na-mic.org
Project Status
• Fully implemented in ITK code thanks to the Programming Week
• Currently writing a paper for the Insight journal detailing the open source nature of the ITK code (i.e. was able to use code from 14 separate ITK filters)
• Finishing touches still in progress
Step 1
![Page 10: A Plan for Brain Connectivity Analysis John Melonakos](https://reader035.vdocuments.mx/reader035/viewer/2022081519/56813ad2550346895da30039/html5/thumbnails/10.jpg)
10National Alliance for Medical Image Computing http://na-mic.org
Some Pictures
Raw Result
Step 1
![Page 11: A Plan for Brain Connectivity Analysis John Melonakos](https://reader035.vdocuments.mx/reader035/viewer/2022081519/56813ad2550346895da30039/html5/thumbnails/11.jpg)
11National Alliance for Medical Image Computing http://na-mic.org
STEP 2: Subdivide GM
Step 2
• Work with Jim Fallon @ UCI (Core 3)
![Page 12: A Plan for Brain Connectivity Analysis John Melonakos](https://reader035.vdocuments.mx/reader035/viewer/2022081519/56813ad2550346895da30039/html5/thumbnails/12.jpg)
12National Alliance for Medical Image Computing http://na-mic.org
More Sketches
Step 2
![Page 13: A Plan for Brain Connectivity Analysis John Melonakos](https://reader035.vdocuments.mx/reader035/viewer/2022081519/56813ad2550346895da30039/html5/thumbnails/13.jpg)
13National Alliance for Medical Image Computing http://na-mic.org
Semi-Automated
• Work with Ramsey Al-Hakim on DLPFC Slicer project– Writing code to wrap the ITK Bayesian
filter in VTK for use in our DLPFC Slicer Module
Step 2
![Page 14: A Plan for Brain Connectivity Analysis John Melonakos](https://reader035.vdocuments.mx/reader035/viewer/2022081519/56813ad2550346895da30039/html5/thumbnails/14.jpg)
14National Alliance for Medical Image Computing http://na-mic.org
STEP 3: DTI Fibers
• Work with Eric and Xavier
Step 3
dss ),( )C(
![Page 15: A Plan for Brain Connectivity Analysis John Melonakos](https://reader035.vdocuments.mx/reader035/viewer/2022081519/56813ad2550346895da30039/html5/thumbnails/15.jpg)
15National Alliance for Medical Image Computing http://na-mic.org
DTI: Artistic Rendition
Step 3
![Page 16: A Plan for Brain Connectivity Analysis John Melonakos](https://reader035.vdocuments.mx/reader035/viewer/2022081519/56813ad2550346895da30039/html5/thumbnails/16.jpg)
16National Alliance for Medical Image Computing http://na-mic.org
DTI: More Art
Step 3
![Page 17: A Plan for Brain Connectivity Analysis John Melonakos](https://reader035.vdocuments.mx/reader035/viewer/2022081519/56813ad2550346895da30039/html5/thumbnails/17.jpg)
17National Alliance for Medical Image Computing http://na-mic.org
The Centrum Ovale Problem
Step 3
![Page 18: A Plan for Brain Connectivity Analysis John Melonakos](https://reader035.vdocuments.mx/reader035/viewer/2022081519/56813ad2550346895da30039/html5/thumbnails/18.jpg)
18National Alliance for Medical Image Computing http://na-mic.org
DTI Reading
Selected Readings
• Eric Pichon’s HBJ approach
• Basser & LiBihan’s early tensor work
• Dave Tuch’s Q-Ball work
• Isabelle Corouge’s DTI shape models
• and more … (currently taking suggestions)
Step 3
![Page 19: A Plan for Brain Connectivity Analysis John Melonakos](https://reader035.vdocuments.mx/reader035/viewer/2022081519/56813ad2550346895da30039/html5/thumbnails/19.jpg)
19National Alliance for Medical Image Computing http://na-mic.org
Questions?