ECE 6357 Advanced Image Processing
A graduate course on the mathematical and algorithmic foundations of (medical) image analysis, from statistical intensity models and segmentation to image registration.
The course works through primary literature and hands-on implementation: translating ideas into mathematical models, implementing them correctly, and understanding why and when they work, or break.
Co-taught with Yihao Liu.
Route through the atlas
- Ch 2.1
- Ch 2.2
Multilevel thresholds planned
Reddi, Rudin and Keshavan, 1984
- Ch 2.3
k-means
When each sample carries several numbers, how do we split the samples into groups?
MacQueen, 1967
- Ch 2.4
- Ch 2.5
Mixtures and EM planned
Dempster, Laird and Rubin, 1977
- Ch 2.6
Bias field correction planned
Van Leemput et al., 1999
- Ch 3.1
- Ch 3.2
Snakes
Given scattered evidence of edges, how do we find one closed boundary?
Kass, Witkin and Terzopoulos, 1988
- Ch 3.3
Gradient vector flow
When the snake cannot feel the boundary, can a better force reach it?
Xu and Prince, 1998
- Ch 3.4
Level sets
How can one boundary become two, without anyone deciding where to cut?
- Ch 3.5
Introduction to Modern Techniques for Medical Image Segmentation seminar
Dr. Dingjie Su, Vanderbilt University
Wednesday, October 21, 2026, 11:15 am – 12:05 pm, Central Time · Engineering & Science Building, Room 001
- O. Ronneberger, P. Fischer and T. Brox, “U-Net: convolutional networks for biomedical image segmentation,” MICCAI, 2015.
- F. Isensee et al., “nnU-Net: a self-configuring method for deep learning-based biomedical image segmentation,” Nature Methods, 2021.
- A. Kirillov et al., “Segment anything,” ICCV, 2023.
- Ch 3.6
Segmentation Evaluation for Medical Image Processing seminar
Dr. Aaron Carass, Johns Hopkins University
Monday, October 26, 2026, 11:15 am – 12:05 pm, Central Time · Engineering & Science Building, Room 001
- Ch 4.1
Transformation and interpolation
An image has values only at its pixels. How do we move it, when every moved pixel lands between them?
Thévenaz, Blu and Unser, 2000
- Ch 4.2
Image similarity
To line up two scans, a computer needs one number for how well they match. What should that number be, when even the brightness of the same tissue differs between the scans?
Maes, Collignon et al.; Viola and Wells, 1997
- Ch 4.3
Linear registration
We can score how well two scans match for any rotation, shift or stretch. With far too many candidates to try them all, how do we find the best one?
Jenkinson and Smith, 2001
- Ch 4.4
Optical flow planned
Horn and Schunck; Lucas and Kanade, 1981
- Ch 4.5
Demons planned
Thirion, 1998
- Ch 4.6
Modern registration seminar
- B. B. Avants et al., “Symmetric diffeomorphic image registration with cross-correlation (SyN),” Medical Image Analysis, 2008.
- G. Balakrishnan et al., “VoxelMorph: a learning framework for deformable medical image registration,” IEEE TMI, 2019.
- J. Chen et al., “TransMorph: Transformer for unsupervised medical image registration,” Medical Image Analysis, 2022.
- Y. Liu et al., “Vector field attention for deformable image registration,” Journal of Medical Imaging, 2024.
- Ch 4.7
Evaluating registration seminar
- Ch 5
Image Analysis in Practice: Modern Tools and Quantitative Modeling for Medical Imaging seminar
Dr. Michael Kim, Columbia University
Friday, November 6, 2026, 11:15 am – 12:05 pm, Central Time · Engineering & Science Building, Room 001
Modern tools and quantitative modeling, 2026
See it on the atlas.