Use this URL to cite or link to this record in EThOS:
Title: Temporal case-based reasoning for insulin decision support
Author: Brown, Daniel
ISNI:       0000 0004 7230 5890
Awarding Body: Oxford Brookes University
Current Institution: Oxford Brookes University
Date of Award: 2015
Availability of Full Text:
Access from EThOS:
Access from Institution:
Type 1 diabetes mellitus is an autoimmune disease resulting in insufficient insulin to regulate blood glucose levels. The condition can be successfully managed through effective blood glucose control, one aspect of which is the administration of bolus insulin. Formulas exist to estimate the required bolus, and have been adopted by existing mobile expert systems. These formulas are shown to be effective but are unable to automatically adapt to an individual. This research resolves the limitations of existing formula based calculators by using case-based reasoning to automatically improve bolus advice. Case-based reasoning is a method of artificial intelligence that has successfully been adopted in the diabetes domain previously, but has primarily been limited to assisting doctors with therapy adjustments. Here case-based reasoning is instead used to directly assist the patient. The case-based reasoning process is enhanced for bolus advice through a temporal retrieval algorithm coupled with domain specic automated adjustment and revision. This temporal retrieval algorithm includes factors from previous events to improve the prediction of a bolus dose. The automated adjustment then refines the predicted bolus dose, and automated revision improves the prediction for future advice through the evaluation of the resulting blood glucose level. Analysis of the temporal retrieval algorithm found that it is capable of predicting bolus advice comparable to closed-loop simulation and existing formulas, with adapted advice resulting in improvements to simulated blood glucose control. The learning potential of the model is made evident through further improvements in blood glucose control when using revised advice. The system is implemented on a mobile device with a focus on safety using formal methods to help ensure actions performed do not violate the system constraints. Performance analysis demonstrated acceptable response times, providing evidence that this approach is viable. The research demonstrates how formula based mobile bolus calculators can be replaced by an artificially intelligent alternative which continuously learns to improve advice.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available