Show simple item record

dc.contributor.advisorOelfke, U
dc.contributor.authorKieselmann, J
dc.contributor.editorOelfke, U
dc.date.accessioned2019-11-14T16:22:22Z
dc.date.issued2019-09-30
dc.identifier.citation2019
dc.identifier.urihttps://repository.icr.ac.uk/handle/internal/3414
dc.description.abstractThe dramatic increase of magnetic resonance imaging (MRI) in daily treatment planning and response assessment of radiotherapy (RT) requires the development of reliable auto-segmentation algorithms for organs-at-risk (OARs) and radiation targets. The current practice of manual segmentation is subjective and time-consuming, particularly for head and neck cancer (HNC) patients. New methodologies based on machine learning offer ample opportunities to solve this problem. This thesis aimed to develop accurate and rapid auto-segmentation algorithms on MR images of HNC patients, employing established atlas-based algorithms and comparing the results with deep learning-based methods. The work is divided into design and implementation of auto-segmentation methods followed by extensive validation studies. For the latter, I developed a fully automated RT workflow enabling validation on purely geometric features of the automatically generated contours whose impact on key dosimetric features of a treatment plan was further analysed. A common challenge for medical image segmentation is the limited availability of data due to the associated cost of obtaining expert contours. Moreover, frequent updates of imaging protocols or scanners may prevent algorithms, developed on existing databases, from working well on newly-acquired images. I designed domain adaptation methods which leverage large databases from related application domains to tackle this problem. While both auto-segmentation strategies achieved clinically acceptable accuracy, atlas-based methods were slow and are, unlike deep learning-based models, difficult to share between hospitals due to data-confidentiality issues. Deep learning-based methods were able to alleviate the computational burden, generating contours within seconds. Moreover, when healthy tissue was infiltrated with irregular structures, deep learning was more accurate. In conclusion, I demonstrated that auto-segmentation was feasible and can change clinical practice. Moreover, domain adaptation strategies hold promise in mitigating problems with small datasets in medical imaging and in eliminating the need to acquire new annotated datasets for each change in imaging protocols.
dc.languageeng
dc.language.isoeng
dc.publisherInstitute of Cancer Research (University Of London)
dc.rights.urihttps://www.rioxx.net/licenses/all-rights-reserved
dc.subjectTheses, Doctoral
dc.subjectHead and Neck Cancer
dc.subjectMagnetic Resonance Imaging
dc.titleNovel concepts for automated segmentation to facilitate MRI-guided radiotherapy in head and neck cancer
dc.typeThesis or Dissertation
dcterms.accessRightsPublic
dcterms.licensehttps://www.rioxx.net/licenses/all-rights-reserved
rioxxterms.licenseref.urihttps://www.rioxx.net/licenses/all-rights-reserved
rioxxterms.licenseref.startdate2019-09-30
rioxxterms.typeThesis
pubs.notes6 months
pubs.organisational-group/ICR
pubs.organisational-group/ICR/Primary Group
pubs.organisational-group/ICR/Primary Group/ICR Divisions
pubs.organisational-group/ICR/Primary Group/ICR Divisions/Radiotherapy and Imaging
pubs.organisational-group/ICR/Primary Group/ICR Divisions/Radiotherapy and Imaging/Radiotherapy Physics Modelling
pubs.embargo.terms6 months
icr.researchteamRadiotherapy Physics Modellingen_US
dc.contributor.icrauthorKieselmann, Jenniferen
uketdterms.institutionInstitute of Cancer Research
uketdterms.qualificationlevelDoctoral
uketdterms.qualificationnamePh.D
dc.type.qualificationlevelDoctoral
dc.type.qualificationnamePh.D


Files in this item

Thumbnail

This item appears in the following collection(s)

Show simple item record