Motion Estimation from Temporally and Spatially Sparse Medical Image Sequences – Niklas Gunnarsson

Temporal medical image sequence, so-called intra-interventional medical images, aids professionals with diagnosis and guidance during ongoing treatment sessions. Fast motions caused by, for example, respiration or heartbeats, set high demands on the imaging procedure and often result in a trade-off between the spatial and temporal resolution of the images.
This talk starts with an example from the radiotherapy domain where the temporal resolution is 200 ms, which limits the spatial resolution to a 2D slice through the 3D volume. Given 2D image observations, I will derive methods to estimate the full 4D motion (3D+t). The methods combine dynamical modeling techniques and medical image registration using traditional and recently developed deep learning methods, like diffusion models.

Keywords: Medical image registration, dynamic modeling, deep learning