Approaches to knowledge-light adaptation in case-based reasoning for radiotherapy treatment planning

Khussainova, Gulmira (2016) Approaches to knowledge-light adaptation in case-based reasoning for radiotherapy treatment planning. PhD thesis, University of Nottingham.

[img] PDF (Thesis - as examined) - Repository staff only - Requires a PDF viewer such as GSview, Xpdf or Adobe Acrobat Reader
Download (3MB)

Abstract

In radiotherapy, ionised radiation beams are used to destroy cancerous cells. A radiotherapy treatment plan needs to be created to deliver a sufficient radiation dose to cancerous cells while sparing nearby organs at risk and healthy tissue. The development of such a treatment plan is a time consuming trial and error process which can take from a few hours up to a few days.

This thesis builds on the previously developed Case-Based Reasoning (CBR) system for radiotherapy treatment planning for brain cancer that was developed in collaboration with Nottingham University Hospitals NHS Trust, City Hospital Campus, UK. The original CBR system focused on the retrieval stage of CBR, where the most similar case was retrieved for the new patient case. The results obtained were promising but adaptation needed to be performed for them to be suitable for the new patient.

Testing of the CBR system by medical physicists has revealed that some of the retrieved radiation beams were not suitable for the tumour position of the new cases and thus could not be used. To avoid this the clustering of cases by their tumour positions was implemented to only retrieve cases with similar tumour positions. The revised CBR system should now retrieve treatment plans with better suited beams.

Adaptation requires a lot of domain knowledge which is often difficult to acquire. In this research we present adaptation approaches which are knowledge-light, i.e. they utilise knowledge available in the case base without requiring interaction with medical experts. Adaptation methods based on machine learning algorithms, in particular neural networks, the naive Bayes classifier, and support vector machines, were developed. Also, an adaptation-guided retrieval approach is presented, in which the case is retrieved only if it can be adapted. In addition, a pair of similar cases are retrieved with it, which guide the adaptation process. The developed knowledge-light adaptation methods have improved the results of the original CBR system. In addition, the proposed adaptation methods are general and could be used in domains where the available amount of knowledge is limited.

Item Type: Thesis (University of Nottingham only) (PhD)
Supervisors: Petrovic, Sanja
Jagannathan, Rupa
Subjects: Q Science > Q Science (General)
Faculties/Schools: UK Campuses > Faculty of Social Sciences, Law and Education > Nottingham University Business School
Item ID: 33110
Depositing User: Khussainova, Gulmira
Date Deposited: 18 Jul 2016 06:40
Last Modified: 13 Sep 2016 22:59
URI: http://eprints.nottingham.ac.uk/id/eprint/33110

Actions (Archive Staff Only)

Edit View Edit View