Use this URL to cite or link to this record in EThOS: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.598263
Title: Computing models from 3D ultrasound
Author: Dance, C. R.
Awarding Body: University of Cambridge
Current Institution: University of Cambridge
Date of Award: 1997
Availability of Full Text:
Full text unavailable from EThOS.
Please contact the current institution’s library for further details.
Abstract:
The computation of geometric models is a central task in imaging science, but it has only been attempted recently for 3D ultrasound. This thesis considers the case of 3D free-hand ultrasound in which a position sensor is attached to a conventional 2D probe. This is an inexpensive, safe and portable technique. However, ultrasound images typically have lower resolutions than images obtained by other clinical modalities and they contain spatially correlated speckle noise. Furthermore, the position and orientation information available is non-ideal and image planes may be irregularly located. Model computation is decomposed into three stages: segmentation, in which the boundaries of organs are located on individual 2D images, registration, in which the boundaries from different scans are aligned to compensate for sensor imperfections and organ motions, and reconstruction, in which 2D boundaries are combined into 3D models. Segmentation is performed by active contours. These enable the use of prior information about the shape of an organ boundary and can therefore be made reliable for noisy images. The aim has been to generate a fast, repeatable and easy to use semi-automated segmentation system. This has been achieved by incorporating a novel method for the propagation of trial active contour shapes between frames of a 3D ultrasound sequence. An intuitive graphical user interface has also been an important part of the design. Segmentations of real scan sequences are discussed and the performance of the scheme is closely evaluated. Registration is achieved by matching points from contours obtained through segmentation. Each image plane is moved by a rigid body transform and all planes are moved simultaneously. This is more efficient than alignment directly from images since it considers few correspondences. It is also more accurate than iterative approaches which align each plane in turn with previously moved planes. Novel matching schemes and means of moving planes are proposed. Regularisation by a model of likely organ motions is necessary since the problem is ill-posed; without such a model, registered contours can less closely resemble the true organ than unregistered ones. Registrations of synthetic and real contours are analysed.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID: uk.bl.ethos.598263  DOI: Not available
Share: