Presentation Type
Poster
Hierarchical and Spatial Analysis of the Movement Aberration Patterns for the Estimation of Cognitive Functions
Abstract
Dementia and Alzheimer’s disease are incurable. There are few state of the art sensor based approaches to quantitatively measure the severity of the patient’s cognitive function. This presented research developed a novel multiscale spatial data analysis approach to quantify the cognitive function, which will lead to a deeper understanding of the development of mental disorders. Spatial location wireless sensors, i.e., accelerometers, were attached to the wrists of fourteen elderly residents for recording movements over a period of thirty days. We have extracted a total of 38 features by the statistical analysis and multi-scale analysis of 3-dimensional spatial data. These quantifiers were then analyzed using best subset regression approach to develop a parsimony model that efficiently predicts each patient’s folstein score. Our results show that a multiple linear regression model with only 5 quantifiers can yield over 95% prediction accuracy. The model was further validated using residual analysis and Bayesian information criteria. We will test the developed regression model on a separate group of patients in the short future.
Categories
Engineering/Physical Science
Research Type
Research Assistant
Mentor Information
Dr. Hui Yang
Hierarchical and Spatial Analysis of the Movement Aberration Patterns for the Estimation of Cognitive Functions
Dementia and Alzheimer’s disease are incurable. There are few state of the art sensor based approaches to quantitatively measure the severity of the patient’s cognitive function. This presented research developed a novel multiscale spatial data analysis approach to quantify the cognitive function, which will lead to a deeper understanding of the development of mental disorders. Spatial location wireless sensors, i.e., accelerometers, were attached to the wrists of fourteen elderly residents for recording movements over a period of thirty days. We have extracted a total of 38 features by the statistical analysis and multi-scale analysis of 3-dimensional spatial data. These quantifiers were then analyzed using best subset regression approach to develop a parsimony model that efficiently predicts each patient’s folstein score. Our results show that a multiple linear regression model with only 5 quantifiers can yield over 95% prediction accuracy. The model was further validated using residual analysis and Bayesian information criteria. We will test the developed regression model on a separate group of patients in the short future.
