Automatic 3D segmentation of the prostate on magnetic resonance images for radiotherapy planning
Abstract. Accurate segmentation of the prostate, the seminal vesicles, the bladder and the rectum is a crucial step for planning radiotherapy (RT) procedures. Modern radiotherapy protocols have included the delineation of the pelvic organs in magnetic resonance images (MRI), as the guide to the ther...
- Autores:
-
Alvarez Jiménez, Charlems
- Tipo de recurso:
- Fecha de publicación:
- 2015
- Institución:
- Universidad Nacional de Colombia
- Repositorio:
- Universidad Nacional de Colombia
- Idioma:
- spa
- OAI Identifier:
- oai:repositorio.unal.edu.co:unal/54354
- Acceso en línea:
- https://repositorio.unal.edu.co/handle/unal/54354
http://bdigital.unal.edu.co/49258/
- Palabra clave:
- 53 Física / Physics
6 Tecnología (ciencias aplicadas) / Technology
61 Ciencias médicas; Medicina / Medicine and health
62 Ingeniería y operaciones afines / Engineering
Radiotherapy planning
MRI prostate segmentation
Atlas based approaches
Label fusion strategy
Planeación de la radioterapia
Segmentación de la próstata en imágenes de resonancia magnetica
Enfoques basados en atlas
Estrategía de fusión de etiquetas
- Rights
- openAccess
- License
- Atribución-NoComercial 4.0 Internacional
| Summary: | Abstract. Accurate segmentation of the prostate, the seminal vesicles, the bladder and the rectum is a crucial step for planning radiotherapy (RT) procedures. Modern radiotherapy protocols have included the delineation of the pelvic organs in magnetic resonance images (MRI), as the guide to the therapeutic beam irradiation over the target organ. However, this task is highly inter and intra-expert variable and may take about 20 minutes per patient, even for trained experts, constituting an important burden in most radiological services. Automatic or semi-automatic segmentation strategies might then improve the efficiency by decreasing the measured times while conserving the required accuracy. This thesis presents a fully automatic prostate segmentation framework that selects the most similar prostates w.r.t. a test prostate image and combines them to estimate the segmentation for the test prostate. A robust multi-scale analysis establishes the set of most similar prostates from a database, independently of the acquisition protocol. Those prostates are then non-rigidly registered towards the test image and fusioned by a linear combination. The proposed approach was evaluated using a MRI public dataset of patients with benign hyperplasia or cancer, following different acquisition protocols, namely 26 endorectal and 24 external. Evaluating under a leave-one-out scheme, results show reliable segmentations, obtaining an average dice coefficient of 79%, when comparing with the expert manual segmentation. |
|---|