2020/11/03 by Julian Krebs, Hervé Delingette, Krebs, Julian +5
Computer Science · Medicine · #Advanced MRI Techniques and Applications #Advanced Neuroimaging Techniques and Applications #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #Generative Adversarial Networks and Image Synthesis #Medical Image Segmentation Techniques
paper · pdf · doi:10.48550/arxiv.2011.01741
openalex publication_date 2020/11/03 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
We propose to learn a probabilistic motion model from a sequence of images\nfor spatio-temporal registration. Our model encodes motion in a low-dimensional\nprobabilistic space - the motion matrix - which enables various motion analysis\ntasks such as simulation and interpolation of realistic motion patterns\nallowing for faster data acquisition and data augmentation. More precisely, the\nmotion matrix allows to transport the recovered motion from one subject to\nanother simulating for example a pathological motion in a healthy subject\nwithout the need for inter-subject registration. The method is based on a\nconditional latent variable model that is trained using amortized variational\ninference. This unsupervised generative model follows a novel multivariate\nGaussian process prior and is applied within a temporal convolutional network\nwhich leads to a diffeomorphic motion model. Temporal consistency and\ngeneralizability is further improved by applying a temporal dropout training\nscheme. Applied to cardiac cine-MRI sequences, we show improved registration\naccuracy and spatio-temporally smoother deformations compared to three\nstate-of-the-art registration algorithms. Besides, we demonstrate the model's\napplicability for motion analysis, simulation and super-resolution by an\nimproved motion reconstruction from sequences with missing frames compared to\nlinear and cubic interpolation.\n