![Page 2: Applied Probability Lecture 5 Tina Kapur tkapur@ai.mit.edu](https://reader035.vdocuments.site/reader035/viewer/2022072006/56649cfa5503460f949cbe7b/html5/thumbnails/2.jpg)
Review Timeline/Administrivia
• Friday: vocabulary, Matlab• Monday: start medical segmentation project• Tuesday: complete project• Wednesday: 10am exam• Lecture: 10am-11am, Lab: 11am-12:30pm• Homework (matlab programs):
– PS 4: due 10am Monday
– PS 5(project): due 12:30pm Tuesday
![Page 3: Applied Probability Lecture 5 Tina Kapur tkapur@ai.mit.edu](https://reader035.vdocuments.site/reader035/viewer/2022072006/56649cfa5503460f949cbe7b/html5/thumbnails/3.jpg)
Review Friday: Vocabulary
• Random variable
• Discrete vs. continuous random variable
• Uniform PDF
• Gaussian PDF
• Bayes rule / Conditional probability
• Marginal Probability
![Page 4: Applied Probability Lecture 5 Tina Kapur tkapur@ai.mit.edu](https://reader035.vdocuments.site/reader035/viewer/2022072006/56649cfa5503460f949cbe7b/html5/thumbnails/4.jpg)
Gaussian PDF
2var
22
2)(
2
1)(
iance
mean
x
exP
![Page 5: Applied Probability Lecture 5 Tina Kapur tkapur@ai.mit.edu](https://reader035.vdocuments.site/reader035/viewer/2022072006/56649cfa5503460f949cbe7b/html5/thumbnails/5.jpg)
Objective
Probabilistic Segmentation of MRI images.
![Page 6: Applied Probability Lecture 5 Tina Kapur tkapur@ai.mit.edu](https://reader035.vdocuments.site/reader035/viewer/2022072006/56649cfa5503460f949cbe7b/html5/thumbnails/6.jpg)
Objective
Probabilistic Segmentation of MRI images.
![Page 7: Applied Probability Lecture 5 Tina Kapur tkapur@ai.mit.edu](https://reader035.vdocuments.site/reader035/viewer/2022072006/56649cfa5503460f949cbe7b/html5/thumbnails/7.jpg)
Today and Tomorrow
• Lecture: Bayesian Segmentation of MRI– Inputs and outputs– Mechanics
• Lab/Recitation: Implementation using Matlab.
![Page 8: Applied Probability Lecture 5 Tina Kapur tkapur@ai.mit.edu](https://reader035.vdocuments.site/reader035/viewer/2022072006/56649cfa5503460f949cbe7b/html5/thumbnails/8.jpg)
Bayesian Segmentation of MRI: Inputs and Outputs
![Page 9: Applied Probability Lecture 5 Tina Kapur tkapur@ai.mit.edu](https://reader035.vdocuments.site/reader035/viewer/2022072006/56649cfa5503460f949cbe7b/html5/thumbnails/9.jpg)
Bayesian Segmentation of MRI: Inputs and Outputs
• Input: 256x256 MRI image
![Page 10: Applied Probability Lecture 5 Tina Kapur tkapur@ai.mit.edu](https://reader035.vdocuments.site/reader035/viewer/2022072006/56649cfa5503460f949cbe7b/html5/thumbnails/10.jpg)
Bayesian Segmentation of MRI: Inputs and Outputs
• Input: 256x256 MRI image
• Given knowledge base– # classes 3: WM (1), GM (2), CSF(3)– Training data (manual segmentations)
![Page 11: Applied Probability Lecture 5 Tina Kapur tkapur@ai.mit.edu](https://reader035.vdocuments.site/reader035/viewer/2022072006/56649cfa5503460f949cbe7b/html5/thumbnails/11.jpg)
Bayesian Segmentation of MRI: Inputs and Outputs
• Input: 256x256 MRI image
• Given knowledge base– # classes 3: WM (1), GM (2), CSF(3)– Training data (manual segmentations)
• Output: segmented image with labels 1,2,3.
![Page 12: Applied Probability Lecture 5 Tina Kapur tkapur@ai.mit.edu](https://reader035.vdocuments.site/reader035/viewer/2022072006/56649cfa5503460f949cbe7b/html5/thumbnails/12.jpg)
Bayes Rule for Segmentation
)(
)()|()|(
BP
APABPBAP
![Page 13: Applied Probability Lecture 5 Tina Kapur tkapur@ai.mit.edu](https://reader035.vdocuments.site/reader035/viewer/2022072006/56649cfa5503460f949cbe7b/html5/thumbnails/13.jpg)
Bayes Rule for Segmentation
)(
)()|()|(
BP
APABPBAP
i
PxPx )()|()P(
:x ofy probabilit marginal the is P(x) where
ii
i
PxP
PxP
x
PxPx
)()|(
)()|(
)P(
)()|()|P(
ii
ii
iii
In MRI Segmentation:
![Page 14: Applied Probability Lecture 5 Tina Kapur tkapur@ai.mit.edu](https://reader035.vdocuments.site/reader035/viewer/2022072006/56649cfa5503460f949cbe7b/html5/thumbnails/14.jpg)
Bayes Rule for Segmentation
)(
)()|()|(
BP
APABPBAP
i
PxPx )()|()P(
:x ofy probabilit marginal the is P(x) where
ii
i
PxP
PxP
x
PxPx
)()|(
)()|(
)P(
)()|()|P(
ii
ii
iii
In MRI Segmentation:
What is P(x|PP(x
![Page 15: Applied Probability Lecture 5 Tina Kapur tkapur@ai.mit.edu](https://reader035.vdocuments.site/reader035/viewer/2022072006/56649cfa5503460f949cbe7b/html5/thumbnails/15.jpg)
Bayesian Segmentation of MRI: Mechanics
• Create class-conditional Gaussian density models from training data
![Page 16: Applied Probability Lecture 5 Tina Kapur tkapur@ai.mit.edu](https://reader035.vdocuments.site/reader035/viewer/2022072006/56649cfa5503460f949cbe7b/html5/thumbnails/16.jpg)
Bayesian Segmentation of MRI: Mechanics
• Create class-conditional Gaussian density models from training data
• Use Uniform priors on the classes
![Page 17: Applied Probability Lecture 5 Tina Kapur tkapur@ai.mit.edu](https://reader035.vdocuments.site/reader035/viewer/2022072006/56649cfa5503460f949cbe7b/html5/thumbnails/17.jpg)
Bayesian Segmentation of MRI: Mechanics
• Create class-conditional Gaussian density models from training data
• Use Uniform priors on the classes
• Use Bayes rule to compute Posterior probabilities for each class
![Page 18: Applied Probability Lecture 5 Tina Kapur tkapur@ai.mit.edu](https://reader035.vdocuments.site/reader035/viewer/2022072006/56649cfa5503460f949cbe7b/html5/thumbnails/18.jpg)
Bayesian Segmentation of MRI: Mechanics
• Create class-conditional Gaussian density models from training data
• Use Uniform priors on the classes
• Use Bayes rule to compute Posterior probabilities for each class
• Assign label of M-A-P class => segmentation
![Page 19: Applied Probability Lecture 5 Tina Kapur tkapur@ai.mit.edu](https://reader035.vdocuments.site/reader035/viewer/2022072006/56649cfa5503460f949cbe7b/html5/thumbnails/19.jpg)
Recitation/Lab
• Start MRI Segmentation Lab