Estimation of Dementia Severity Using SVM based on Patient's Engagement Levels in Conversation

Abstract

In this study, we aimed to investigate how robotic media can promote the conversation of older adults with dementia and their engagement in conversation can be helpful for assessing dementia severity. Our target group was patients diagnosed with dementia by a specialist and a teleoperated android robot was applied to extract their responses. Based on video recordings, participants' excitement levels during conversation were observed and scored by an experimenter. To predict participants' dementia severity, the excitement level was counted as a variable together with participants' basic information such as age, gender, dementia types, years of education and duration of disease. Participants were divided into two dementia severity groups according to the Clinical Dementia Rating (CDR) scale: mild (group 1), and moderate or severe (group 2) where in the latter the severity of dementia symptoms called BPSD is known to be the highest. We trained a support vector machine to predict the two groups of CDR scores, and the classifier exhibited the highest prediction accuracy of 80% with the variables of the excitement level and duration of disease. The result suggests the potential of predicting dementia severity from the patient's engagement in daily conversation and his or her basic information. The simple method may have a role in the early detection of BPSD. Thus, we further discuss key issues identified in this study and how the patient's daily responses to a robot needs to be utilized for managing BPSD.