Effect of family income on the children's hippocampus,
which regulates memory: example of longitudinal MRI study.
Lectures
Tuesday/Thursday 2:30-3:45pm
Engineering Hall 2535
Instructors
Moo K. Chung
mkchung@wisc.edu
Jeanette A. Mumford jeanette.mumford@gmail.com
Department of Biostatistics and Medical Informatics
Waisman Laboratory for Brain Imaging and Behavior
University of Wisconsin-Madison
Office Hours
Moo Chung: every Thursday1:00-2:00pm (MSC
5785) except during the weeks of January 27-Febuary 3.
Jeanette A.Mumford: Every Wednesday 3:00-4:00pm (Wasiman Center
T233) except during the week of February 23-25.
For other dates, make an appointment with us.
Audience
Undergraduate
and graduate students from biological, medical, and computer
science departments who wish to learn the basics of medical
image analysis. The student enrollment is limited to 25.
Prerequisites
CS 367
(or programming experience with some high-level programming
language), Math 221, and some familiarity with introductory
concepts in probability and linear algebra. or by consent of the
instructor.
Course Objectives
Present
introductory medical image processing and analysis techniques. Basic concepts without
extensive technical details or proofs will be covered. Students and researchers
should gain an easy to follow understanding of various
processing and analytic issues in medical image analysis. The focus will be on
awareness of processing and analysis concepts used in the field
and how some of these ideas can be implemented in Matlab to
facilitate image analysis.
Textbook & Language of Instruction
There is no textbook but reading
materials will be assigned and lecture notes will be distributed
online through the dropbox. Homework will be distributed online.
Link to the Dropbox folder:
MATLAB is the language of instruction. Free MATLAB and other
commercial software packages can be downloaded from https://software.wisc.edu/cgi-bin/ssl/csl.cgi
for Univ. of Wisconsin system students.
Course Evaluation
Students
will be assigned written and programming homework assignments (4
homework 30%), and two quizzes (60%) and final exam or project
(take home 10%). The grading scheme is
as follows: A (above 85%), AB (above 80%), B (above 75%), BC
(above 70%), C (above 65%), D (above 60%) and F (below 60%).
Sample homework problems:
1.Given longitudinally collected CT
images of the head, describe a sequence of processing and
analysis procedures for modeling mandible volume change.
Implement in Matlab a morphological procedure (e.g., open/close
filtering) for correcting topological defects in the volume
computation due, for example, to teeth fillings.
2.Given an existing implementation of
a generalized Hough transform, write code to adapt it for
detecting circular objects in microscopy images. Using training
data, perform evaluations assessing the accuracy of your method.
3. 0% tolerance for plagiarism.
Wikipedia defines it as "plagiarism is the wrongful appropriation
and stealing and publication of another author's language, thoughts,
ideas, or expressions and the representation of them as one's own
original work". This includes computer codes. Students are required
to work independently from each other and expected NOT to work
together for homework and exam problems.
Course Topics/schedule and additional supplementary reading list
Introduction to
Matlab. The basics
on Matlab programming will be covered so that students can
do homework. Reading:Martinez, W.L, Martinez, A.R.
2007. Computational Statistics Handbook with MATLAB –
Chapter 5 Exploratory data analysis, 2nd edition.
Chapman & Hall/CRC. 3 hours
Introduction
to various imaging modalities (EEG, PET, MRI,fMRI, DTI,
CT and microscope images and their simple manipulation. Data storage formats for different imaging
modalities. This may utilize invited guest lectures by
experts in each imaging modality.Reading:
Dhawan, A.P. 2011 Medical image
analysis - Chapter 1 Introduction. John Wiley
& Sons. Nichols, T., Poldrack, R.A., Mumford, J.A.
Handbook of functional MRI data analysis. Dhawan, A.P. 2011 Medical image
analysis - Chapter 2 Image formation. John
Wiley & Sons; Chung, M.K. 2013
Statistical and Computational Methods in Brain Image
Analysis – Chapter 5 Introduction to Brain and
Medical Images, Chapman & Hall/CRC. 3
hours
Basic image
processing and quantification (Integration,
differentiation, topology). High-pass and low-pass filtering.
Spatial smoothing, temporal smoothing, morphological
operations, etc. Simple quality control and validation
methods such as using image filtering for denoising,
kappa index, and image correlations. Reading: Yao,
T.S. Insight into images: principles and practice
for segmentation, registration and image analysis –
Chapter 1 Introduction and Basics. A K Peters/CRC Press.
9 hours
Surface
processing. 3D volume to surface meshes.
Simple mesh processing and manipulation. Reading: Chung,
M.K., Hanson, J.L., Pollak, S.D. 2016. Statistical
analysis on brain surfaces, Handbook of Modern
Statistical Methods: Neuroimaging Data Analysis. 3 hours
Image complexity.
Measure of image complexity,
fractional dimensions in images and signals etc. 3
hours
Image
segmentation methods. Thresholding, intelligent scissors, mean-shift, and
random-walk and graph-based methods. Reading: Yao, T.S.
Insight into images: principles and practice for
segmentation, registration and image analysis – Chapter 1
Introduction and Basics, Chapter
2 Segmentation. A K Peters/CRC Press. 4.5 hours
Quantification of
multiple images. We will
present a simple statistical methodology that is covered in
BMI 541. A simple example is explaining how to construct a
t-statistic at each voxel and how it is affected by image
filtering, outliers, etc. General linear model
(GLM) will be extensively used. Reading: Chung, M.K. 2013 Statistical and Compuational
Methods in Brain Image Analysis – Chapter 3 General Linear
Models, Chapman & Hall/CRC.
Image
registration. Using Procrustes or iterative closest point
transform. Demonstrate use of some advanced image
registration methods implemented in Matlab and used by the
community (e.g., diffeomorphic, mutual information based). Reading: Yao, T.S.
Insight into images: principles and practice for
segmentation, registration and image analysis – Chapter 3
Registration. A K Peters/CRC Press. 3 hours
Image
classification/categorization. In many medical imaging studies, classification
accuracy is used as an alternative quantitative measure
instead of p-value. Will go over basics on linear
discriminant analysis and logistic
regression. Support vector machine. Present simple examples where the classification
accuracy can be computed by hand. Describe precision-recall
and AUC curves. Reading: Flury, B. A. 1997. First course in multivariate
statistics - Chapter 5 Linear Discriminant Analysis,
Chapter 7 Logistic Regression, Springer. 4.5 hours
Interpretation
of p-values and multiple comparisons. For clinically oriented studies,
it is often necessary to interpret the medical imaging
literature where the results are displayed in p-value maps
with multiple comparisons correction (mostly Bonferroni
correction, FDR and random field theory). We will only cover
how to compute them in Matlab, and how to interpret them,
and will not cover theoretical issues. Reading: Genovese, C.R., Lazar, N.A., Nichols, T.E., 2002.
Thresholding of statistical maps in functional
neuroimaging using the false discovery rate. NeuroImage
15, 870–878;Nichols, T., Hayasaka, S. 2003.
Controlling the familywise error rate in functional
neuroimaging: a comparative review. Stat Methods Med Res.
12:419-46. 1.5 hours