Andreas Mang

Date:  24 Jujy 2019

Time: 12:45 - 13:30 (Lunch available from 12:35)

Room: RA 1501 (Ravelijn)

Speaker: Andreas Mang (University of Houston)


Title: Fast algorithms for nonlinear optimal control for diffeomorphic registration


We present efficient algorithms for nonlinear optimal control for diffeomorphic registration. Our contributions are the design of effective numerical methods and fast parallel computational kernels. We consider different optimal control formulations for matching imaging data in two and three dimensions. The problem is as follows: Given two views of the same scene, we seek a spatial transformation $y$ that relates points in one view of an object to its corresponding points in another view of the same object. In our formulation, we invert for a velocity field $v$ that parameterizes the sought-after transformation $y$. Suitable regularity requirements for $v$ ensure that $v$ gives rise to a diffeomorphism $y$, i.e., $y$ is a bijection and has a smooth inverse.

Our solvers are based on state-of-the-art techniques in scientific computing to enable fast convergence and short runtime. We will discuss different building blocks of our solvers. We will showcase results on real and synthetic data to study the rate of convergence, time-to-solution, numerical accuracy, and scalability of our solvers. As a highlight, we will showcase results for a GPU-accelerated implementation that allows us to solve clinically relevant 3D image registration problems to high accuracy in well under 10 seconds on a single GPU.

This is joint work with Robert Azencott, George Biros, Jiwen He, Miriam Mehl, and others.