Modeling Movement Disorders in Parkinson's Disease using Computational Intelligence

Parkinson's is the second most common neurodegenerative disease after Alzheimer's Disease and affects 127,000 people in the UK alone. Providing the most appropriate treatment pathway can prove challenging owing to the difficulty in obtaining an accurate diagnosis; due to its similarity in symptoms with other neurodegenerative diseases, it is estimated that in the United Kingdom around 24% of cases are misdiagnosed by general neurologists. A means of providing an accurate and early diagnosis of Parkinson's Disease would thereby enable a more effective management of the disease, increased quality of life for patients, and reduce costs to the healthcare system.

The work described in this thesis details progress towards this goal by modeling movement disorders in the form of positional data recorded from simple movement tasks, building towards a fully objective diagnostic system without requiring any specialist domain knowledge. This is accomplished by modeling established movement disorder markers using Evolutionary Algorithms to train ensembles, before implementing feature design strategies with both Genetic Programming and Echo State Networks.

The findings of this study make an important contribution to the area of data mining, including: the demonstration that Computational Intelligence-based feature design strategies can be competitive to conventional models using features extracted with expert domain knowledge; a thorough survey of evolutionary ensemble research; and the development of a novel evolutionary ensemble approach comprising traditional single objective Evolutionary Algorithm. Furthermore, an extension to a Genetic Programming feature design strategy for periodic time series is detailed, in addition to demonstrating that Echo State Networks can be directly applied to time series classification as a feature design method. This research was carried out in the context of building an applied diagnostic aid and required developing models with means of indicating the most discriminatory aspects of the sequence data, thereby facilitating inference of the precise mechanics of movement disorders to clinical neurologists.