2023 Volume 11 Pages 10-19
The number of patients with dementia is increasing worldwide and there is a need to improve the efficiency of cognitive function assessment to assist healthcare professionals in the screening process. We investigated the screening performance for cognitive severity by applying machine learning to a clock-drawing test. The 77 elderly subjects were assigned to the dementia group (COG), the mild cognitive impairment group (MCI), and the healthy group (HC). Clock drawing test was measured for 14 drawing features with or without anomalies as a binary score. A discriminant model was constructed using a pairwise method with a support vector machine. We also investigated the drawing features that contribute to the discrimination, using the feature importance of random forests. The classification accuracy was 70% for the COG versus MCI, 76% for the COG versus HC, and 56% for the MCI versus HC. In addition, for COG versus MCI and COG versus HC classification, five or three features were selected, including hand misrepresentation, long and short hand reversal, and deficit in the spatial layout of numbers, respectively. These drawing features could contribute to efficient primary screening for dementia.
