Classification of history-based dementia data with clinical diagnosis

Loi, Woan Yee (2019) Classification of history-based dementia data with clinical diagnosis. Master dissertation/thesis, UTAR.

Preview

PDF Download (1637Kb) | Preview

Abstract

Dementia is a syndrome caused by various brain illness that influences memory, thinking, behaviour and ability of a person to perform everyday activities. In this study, a modified Clinical Dementia Rating (CDR) questionnaire consists of 42 questions in six functional domains was adopted to detect the intraindividual changes from the person’s prior cognitive capability, and the severity of dementia was then quantified according to the CDR scale. The purpose of this study is to establish an ideal predictive model that can classify the dementia severity through a comparison between traditional statistical methods such as Fisher’s linear discriminant analysis (LDA) and multinomial logistic regression (MLR). In addition, k-means clustering is applied to analyse the similarities or discrepancies among the groups in order to identify dementia symptoms for different severity of dementia. The overall classification accuracy from the cross-validated dataset of the LDA reduced model is 91.60% for three severity levels compared to 75.44% for five severity levels. Meanwhile, the MLR reduced models indicate 82.38% and 93.24% of the average accuracy in respect of the 5dementia and 3-dementia levels respectively. Despite merely 2% lower of the projection accuracy between LDA and MLR reduced models of 3-dementia levels, the LDA model contains more questions to decide the impairment status in each domain. As far as the clinical diagnosis is concerned, the LDA 3-dementia levels with 23 variables is a more trustworthy model from the medical perspective as the questions deliver more accurate diagnoses by medical examiners. The CDR questionnaire has better classification powers for moderate dementia and severe dementia cases, but less sensitive in differentiating normal, uncertain dementia and mild dementia cases. Clustering analysis confirmed that the normal-mild group has the strongest cognition and functionality in memory, orientation, judgement & problem solving, community affairs, and home & hobbies domains, followed by moderate dementia and severe dementia. Nevertheless, it presents contradictory results in respect of the personal care category, and hence further investigations are necessary.

Actions (login required)