Use this URL to cite or link to this record in EThOS:
Title: Biomedical applications of belief networks
Author: Cunliffe, Daniel Jonathan
Awarding Body: University of Edinburgh
Current Institution: University of Edinburgh
Date of Award: 1996
Availability of Full Text:
Access from EThOS:
Full text unavailable from EThOS. Please try the link below.
Access from Institution:
Biomedicine is an area in which computers have long been expected to play a significant role. Although many of the early claims have proved unrealistic, computers are gradually becoming accepted in the biomedical, clinical and research environment. Within these application areas, expert systems appear to have met with the most resistance, especially when applied to image interpretations. In order to improve the acceptance of computerised decision support systems it is necessary to provide the information needed to make rational judgements concerning the inferences the system has made. This entails an explanation of what inferences were made, how the inferences were made and how the results of the inferences are to be interpreted. Furthermore there must be a consistent approach to the combining of information from low level computational processes through to high level expert analyses. Until recently ad hoc formalisms were seen as the only tractable approach to reasoning under uncertainty. A review of some of these formalisms suggests that they are less than ideal for the purposes of decision making. Belief networks provide a tractable way of utilising probability theory as an inference formalism by combining the theoretical consistency of probability for inference and decision making, with the ability to use the knowledge of domain experts. The potential of belief networks in biomedical applications has already been recognised and there has been substantial research into the use of belief networks for medical diagnosis and methods for handling large, interconnected networks. In this thesis the use of belief networks is extended to include detailed image model matching to show how, in principle, feature measurement can be undertaken in a fully probabilistic way. The belief networks employed are usually cyclic and have strong influences between adjacent nodes, so new techniques for probabilistic updating based on a model of the matching process have been developed. An inference shell called FLAPNet has been implemented and used to apply the belief network formalism to the tasks of model based image matching and the incremental aggregation of disparate information for diagnosis. The domains of application are fetal ultrasound imaging and cervical screening respectively. It is argued that belief networks combine the necessary quantitative features required of a decision support system with desirable qualitative features that will lead to improved acceptability of expert systems in the biomedical domain.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available