Search

In this thesis, novel Case-Based Reasoning (CBR) methods were developed to be included in CBRDP (Case-Based Reasoning Dose Planner) -an adaptive decision support system for radiotherapy dose planning. CBR is an artificial intelligence methodology which solves new problems by retrieving solutions to previously solved similar problems stored in a case base. The focus of this research is on dose planning for prostate cancer patients. The records of patients successfully treated in the Nottingham University Hospitals NHS Trust, City Hospital Campus, UK, were stored in a case base and were exploited using case-based reasoning for future decision making. After each successful run of the system, a group based Simulated Annealing (SA) algorithm automatically searches for an optimal/near optimal combination of feature weights to be used in the future retrieval process of CBR. A number of research issues associated with the prostate cancer dose planning problem and the use of CBR are addressed including: (a) trade-off between the benefit of delivering a higher dose of radiation to cancer cells and the risk to damage surrounding organs, (b) deciding when and how much to violate the limitations of dose limits imposed to surrounding organs, (c) fusion of knowledge and experience gained over time in treating patients similar to the new one, (d) incorporation of the 5 years Progression Free Probability and success rate in the decision making process and (e) hybridisation of CBR with a novel group based simulated annealing algorithm to update knowledge/experience gained in treating patients over time. The efficiency of the proposed system was validated using real data sets collected from the Nottingham University Hospitals. Experiments based on a leave-one-out strategy demonstrated that for most of the patients, the dose plans generated by our approach are coherent with the dose plans prescribed by an experienced oncologist or even better. This system may play a vital role to assist the oncologist in making a better decision in less time; it incorporates the success rate of previously treated similar patients in the dose planning for a new patient and it can also be used in teaching and training processes. In addition, the developed method is generic in nature and can be used to solve similar non-linear real world complex problems.