Towards the automation of colon segmentation in magnetic resonance images
(English) The study of the colonic volume has strong relevance in gastroenterology and contributes to the diagnosis and research of lowseverity diseases such as constipation, diarrhea, or Irritable Bowel Syndrome. In the context of these affections, the use of invasive techniques or ionizing imaging...
| Autor: | |
|---|---|
| Tipo de recurso: | tesis doctoral |
| Fecha de publicación: | 2024 |
| País: | España |
| Institución: | Universitat Politècnica de Catalunya (UPC) |
| Repositorio: | UPCommons. Portal del coneixement obert de la UPC |
| Idioma: | inglés |
| OAI Identifier: | oai:upcommons.upc.edu:2117/407235 |
| Acceso en línea: | https://hdl.handle.net/2117/407235 https://dx.doi.org/10.5821/dissertation-2117-407235 |
| Access Level: | acceso abierto |
| Palabra clave: | Àrees temàtiques de la UPC::Informàtica Àrees temàtiques de la UPC::Enginyeria biomèdica |
| Sumario: | (English) The study of the colonic volume has strong relevance in gastroenterology and contributes to the diagnosis and research of lowseverity diseases such as constipation, diarrhea, or Irritable Bowel Syndrome. In the context of these affections, the use of invasive techniques or ionizing imaging is not advisable, and the medical experts opted for Magnetic Resonance Imaging (MRI) acquired without contrast administration nor any form of patient preparation to avoid exogenous colon alteration. Specifically, two MRI modalities are of special interest: T2-weighted half-Fourier acquisition single-shot turbo spin echo sequence (T2-HASTE), which allows the segmentation of the colon lumen, and T1-weighted Fat-Sat (T1-FS), where fecal and gas contents can be distinguished. The colon analysis requires the segmentation of the colon both in T1-FS and T2-HASTE images, which is a cumbersome and timeconsuming process when performed manually. Moreover, manual segmentation involves a high degree of uncertainty in the colon boundaries delimitation, which leads to arbitrariness and poor measurement repeatability. This Ph.D. Thesis proposes an end-to-end quasi-automatic framework that comprises all the steps required to accurately segment the colon in T2-HASTE and T1-FS images, allowing the study of its morphology and the colonic content distribution. In order to ease the validation of the automated segmentation results, a tool for the synchronized T1-FS/T2-HASTE visualization has also been designed. T2-HASTE segmentation is organized as a three-stage pipeline. In the first stage, a custom tubularity filter is run to detect colon candidate areas. The specialists provide a list of points along the colon trajectory, which are combined with tubularity information to calculate an estimation of the colon medial path. In the second stage, the colon region of interest is delimited by applying custom segmentation algorithms to detect colon neighboring regions and the fat capsule containing abdominal organs. Finally, within the reduced search space, the segmentation is performed via 3D graph-cuts in a three-stage multigrid approach. T1-FS colon segmentation is based on a non-rigid registration of the T2-HASTE image towards the T1-FS image. The mapping transformation delivered by the registration is used to translate the T2-HASTE colon segmentation to the T1-FS space. A novel Iterative Colon Registration process corrects the registered colon misalignments by employing a mesh deformation approach guided by a probabilistic model. The model provides information about the presence of feces, colonic gas, and the surrounding fat. Each iteration alternates a deformation phase with a shape-preservation phase to counterbalance the lack of boundary information that characterizes the regions of the colon containing gas. This process converges into an optimal fit of the colon segmentation in the T1-FS. The experiments have proven the accuracy and usability of the algorithms, which contributed to the integration of MRI image-based colon analysis into the clinical routine. The results imply a substantial step towards a fully automated segmentation, speeding up the acquisition of further data for research and diagnosis purposes while improving the repeatability of the measurements. |
|---|