PROJECT LEADER
AI for the detection of gastrointestinal motility in 3D cine MR
Develop a deep learning method to automatically quantify gastrointestinal motion using dynamic 2D and 3D MRI of the abdomen.
FINISHED
2019
|
2024
RESEARCH AREAS
About project
Dynamic MRI allows detection and quantification of motility in the complete gastrointestinal tract without harm for the patient. This has been demonstrated to be valuable in inflammatory bowel disease and may be valuable in the evaluation of other gastrointestinal pathologies. In spite of possibilities offered by imaging, an automatic, fast, robust and reproducible method for motility quantification, especially exploiting dynamic 3D MR data, is currently lacking. Hence, the goal of this project is to develop a deep learning method to automatically quantify gastrointestinal motion using dynamic 2D and 3D MRI of the abdomen.
Members
Our publications & software
Publications
Journal articlesL. D. Van Harten, J. Stoker; I. Išgum, "Robust deformable image registration using cycle-consistent implicit representations", IEEE Transactions on Medical Imaging, 2024; 43 (2): 784-793.
InproceedingsL. van Harten, R.L.M. van Herten, J. Stoker and I. Isgum, "Deformable Image Registration with Geometry-informed Implicit Neural Representations", Medical Imaging with Deep Learning, 2023; 227: 730-742.
L.D. van Harten, C.S. de Jonge, K.J. Beek, J. Stoker, I. Isgum, "Untangling and segmenting the small intestine in 3D cine-MRI using deep learning", Medical Image Analysis, 2022; 78: 102386.
L. van Harten, C. de Jonge, J. Stoker, I. Isgum, "Untangling the small intestine in 3D cine-MRI using deep stochastic tracking", Medical Imaging with Deep Learning (MIDL 2021), 2021.
Software
Deformable Image Registration with Geometry-informed Implicit Neural Representations
A method that incorporates a tangent coordinate system into INR-based registration. The deformation function is represented in the tangent space of a curve describing anatomical structures that influence the motion in time series data, which yields a space that is aligned with the expected motion. The geometric information is therefore explicitly encoded into the neural representation, reducing the complexity of the optimized deformation function.