Synopsis

In this talk, I will give an overview of important aspects of Cardiac MRI processing by the help of advanced deep learning techniques. Our main clinical focus are Tetralogy of Fallot and Duchenne muscular dystrophy cases. When analyzing heterogenous CMR from multiple sites, automatic interpatient motion comparison requires spatial and temporal alignment of the CMR sequences. I show how to align the volume sequences by self-supervised AI methods, which do not require any expert labels. Cardiac phase-to-phase registration is realized by deformable registration models, that automatically derive a displacement field for calculating cardiac strain values.