Efficient large scale multimodal image registration and AI-supported collaborative analysis
Poster presented as part of the Crick BioImage Analysis Symposium 2023.
Correlative multimodal imaging combines several techniques to maximise the information about an object of interest. The first step is accurate image registration – to find a spatial transformation which best aligns the data in the same coordinate system. Reliable and accurate automated alignment of heterogeneous multimodal biomedical images is still a very challenging problem. Mutual Information (MI) is a trusted objective function for the task of multimodal image registration. MI maximization often outperforms modern deep learning approaches. Unfortunately MI exhibits many local optima, making gradient based methods sensitive to initial alignment. Here we present an approach for fast computation of MI in the Fourier domain. This makes grid-search approaches feasible, delivering state of the art rigid multimodal image registration performance.
Permission has been given by authors to upload to Crick Figshare. Copyright remains with the original authors.