Enhanced Interpretation of the Mini-Mental State Examination

The goal of the research reported in this thesis is to contribute to early and accurate detection of dementia. Early detection of dementia is essential to maximising the effectiveness of treatment against memory loss. This goal is pursued by interpreting the Mini-Mental State Examination (MMSE) in novel ways. The MMSE is the most widely used screening tool for dementia, it is a questionnaire of 30 items. The objectives of the research are as follows: • to reduce the dimensions of the MMSE to the most relevant ones in order to inform a predictive model by using computational methods on a data set of MMSE results, • to construct a model predicting a diagnosis informed by the features extracted from the previous step by applying, comparing and combining traditional and novel modelling methods, • to propose a semantic analysis of the sentence writing question in the MMSE in order to utilise information recorded in MMS examinations which has not been considered previously. Traditional methods of analysis are inadequate for questionnaire data such as the MMSE due to assumptions of normally distributed data. Alternative methods for analysis of discrete data are investigated and a novel method for computing information theoretic measures is proposed. The methods are used to demonstrate that an automated analysis of the MMSE sentence improves the accuracy of differentiating between types of dementia. Finally, models are proposed which integrate the semantic annotations with the MMSE data to derive rules for difficult to distinguish types of dementia.