Medical imaging has been increasingly growing in the practice of medicine today. Acquisitions, whether in two or three dimensions, are used before, during or after an operation. However, the display is not always sufficient, it is often necessary to measure the precise volume of organs. To do this, it is imperative to accurately define the borders of the anatomical structures of interest using a segmentation method. This area of research related to medical imaging is constantly changing and has a large number of different approaches. Some systems use statistical databases such as active contours; others use only the specific data image to be processed as Level-Sets. Depending on the type of acquisition and / or the type of targeted organ, some approaches are preferable to others. However, when structures of interest have complex surfaces or topology that may experience strong inter-individual variation, the approaches proposed in the literature may be ineffective. The potential causes of inefficiency can be attributed either to too many degrees of freedom (no prior knowledge) or too many constraints (a priori knowledge too strict). In both cases, the segmentation result may be false and therefore unusable in clinics. To overcome these limitations, the method proposed in this thesis uses the data to build an in-situ prior model reliable for accurate segmentation of target organs. The model generated allows computing a minimal surface in a small flat space. This minimal surface is the solution of the segmentation problem. The developed method is automatic, multi-modality and applicable to several types of organs.
| Date | 28 Jun 2016 |
|---|
| Original language | French |
|---|
| Awarding Institution | - École de technologie supérieure
|
|---|
| Supervisor | Jacques A. de Guise (Supervisor) & Matthieu Schmittbuhl (Co-supervisor) |
|---|
Chantrel, S. (Author),
De Guise, J. (Supervisor) & Schmittbuhl, M. (Co-supervisor),
28 Jun 2016Student thesis: Doctoral thesis › Doctorate in Engineering: Engineering