Medical Image Analysis

• perform landmark-based, intensity-based, and surface-based image registration.
• decide on the most appropriate similarity measure for specific image registration problems.
• use several linear and non-linear transformation models for image registration.
• implement segmentation algorithms based on generative probabilistic models.
• implement simple neural networks for example-based segmentation.
• explain the mathematical models and optimizers used in medical image registration and segmentation.
• validate the results of automated medical analysis algorithms.
• decide on the most appropriate interpolation schemes.

Landmark-based registration; intensity-based registration (sum-of-squared differences, Mutual Information, principal axis transform); linear and non-linear transformation models (rigid and affine transformations, thin plate splines, B-splines); surface-based registration and segmentation; voxel-based segmentation using generative and discriminative methods (Gaussian mixture models, Markov random field priors, neural networks); optimization (Gauss-Newton, expectation-maximization, stochastic gradient descent); atlases; validation.

02402/02502, It is assumed that the participants have had an introductory statistics course and an introductory image analysis course. Furthermore, working knowledge of Python/Matlab is expected.

Online lectures; computer exercises. The lectures will be shared between DTU and Aalto University, and will be streamed.