Motion analysis in 4D MRI of the small intestine using neural networks

Movement of the small intestine is rather difficult to measure. It can be done by manometry (pressure measurements using sensors on a catheter), but this procedure is very invasive. A non-invasive modality that can offer an alternative is 4D MRI. This involves taking multiple 3D MRI images in quick succession so that they can be played back as a video. While 4D MRI scans for the small intestine have existed for a number of years, the analysis of these sequences was still an unsolved problem. Quantifying and assessing a moving 3D structure is a difficult task for clinicians, which is further complicated by the tight packing and complex folding of the small intestine in the abdomen. The aim of this thesis was to develop methods that enable such analysis in 4D MRI of the small intestine, using various applications of neural networks. The thesis introduces a method for automatically untangling the small intestine into a simpler representation, using a swarm of stochastic neural tracking agents to extract intestinal centerlines. It also introduces multiple methods for image registration with implicit neural representations, enabling motion estimation within the intestines. The localisation and motion estimation methods are combined into a motion analysis pipeline that can characterize the movement in imaged intestinal segments, which has potential applications in both research and in clinical settings.