Moo K. Chung
Department of Biostatistics and Medical Informatics
Waisman Laboratory for Brain Imaging and Behavior
University of Wisconsin-Madison
Office: Waisman Center
Personal Webpage: www.stat.wisc.edu/~mchung
Office Hour: 10:45-11:45AM. Right after class. To set up
separate appointments, email me.
course is designed for graduate students, and
researchers who wish to learn quantitative techniques
in analyzing medical images and set up statistical
models. The course material is applicable to a wide
variety of statistical problems in medical and
Motivations for Course
The demand from
students, staff and faculty is exploding as many are
increasingly involved in collecting various medical
imaging data, but there is no course to train students and
researchers on how to analyze medical images
quantitatively. However, there is no statistics course
focused on imaging data in the campus. Most imaging courses from
other department deal mainly with image acquisition or
processing but do not provide relevant in-depth statistical
content to students.
statistical and other quantitative techniques used in
analyzing various medical images. A concise review of
relevant methodological background will be presented.
Basic concepts of key methods will be developed with
considerable attention to analysis of real medical
maging data of various types and problems. Students
and researchers should gain a deeper understanding of
statistical methods used in medical image analysis.
Course projects will be designed to apply methods
learned from classes to medical images provided by an
instructor or students themselves.
Students are required to submit (1)
research proposal and (2) final research project
report and do an oral presentation at the end of the
semester. Students can use their own medical images
for the final project after consultation with the
instructor. For students without their own data set,
the final project topics and data set will be
provided. Sample A-grade level final project reports
can be found here.
is based on the research proposal (10%), oral presentation
(30%) and final project report (60%).
grading scheme is as follows: A (85%), AB (80%), B (75%),
BC (70%), C (65%), D (60%) and F (below 60%).
text book is Computational
Neuroanatomy: The Methods
written by the instructor. It is
not a required textbook thought but each module of lectures (2-3
lectures) will be based on each chapter of the textbook.
Three hour lectures/week will
contain four parts: 1) presentation of statistical
theory/concepts via PowerPoint, 2) presentation of MATAB
routintes demonstrating the methods, 3) presentation of
real medical imaging applications, 4) open discussion
with students. Lectures given in 2012 are found at here.
MATLAB demonstration done in 2012 class can be found at
will give a concise review of relevant statistical
background needed in advanced medical image analysis.
Students will be introduced to key concepts and methods
used in analyzing various medical images and imaging
related problems. Some of topics coverd in the class
- 3D Data and image
manipulaiton/visualization. R/MATLAB packages.
inference on binary images. logistic regression, discrimiant
- Random fields
theory: excursion probablity. Karhunen-Loeve expansion, multiple comparions.
comparisons: Bonferroni correction. Random field theory.
Permutation tests and false discovery rates (FDR). power analysis under
- Hilbert space
methods: Fourier analysis. Functional-PCA. wavelets.
spherical harmonics. hyperspherical harmonics.
- Mixture models:
Gaussian mixture models, EM-algorithms. voxel-based
- Regression on
images: kernel smoothing, heat kernel smoothing, diffusion
- Statistical infernce
on manifolds: Laplace-Beltrami operator, kernels on manifolds.
- General linear
models (GLM): univariate-GLM and multivariate-GLM.
- Inference on shapes
deformable shaps: tensor-based morphometry (TBM),
deformation-based morphometry (DBM).
- Newtork analysis:
Covariance and inverse covariance matrix estimation.
Sparse regression and sparse-PCA. Brain and biological
- Topological Data
analysis: persistent homology, Rips filtrations,
dendrogams and hierarchical clustering.
statistics: Monte-Carlo simulations,
Gibbs sampling, MCMC, bootstrap, jackknife