Appraising cognitive status in dementia via touch-based reaction time: a preliminary machine learning study

Abstract

People with dementia (PwD) perform cognitive-based therapeutic activities. Literature reports a variety of studies exploring relationships between the cognitive status of PwD as determined by the Mini-Mental State Examination (MMSE) and their reaction times from a myriad of stimuli incorporated into cognitive activities. Nevertheless, these technology-supported activities usually include distracting elements, complex instructions, and unfamiliar devices for older adults, introducing bias into reaction times. The objective of this work is to appraise the cognitive status of people with dementia using reaction times from touch interaction tasks. For this purpose, a relatively simple cognitive activity (involving the intuitive tap gesture) and a 32-inch wide touchscreen were designed and implemented. Afterward, 21 PwD from a day center located in Sonora, Mexico were recruited. The participants were instructed to carry out a cognitive activity consisting of five consecutive taps and their reaction times were recorded. The collected data was analyzed using (i) a correlation analysis, (ii) a bootstrap evaluation of machine learning classification models, and (iii) a logistic regression analysis. From the empirical results, it can be concluded that there is a negative relationship between the MMSE score of PwD and the reaction times from taps. In addition, the bootstrapped mean accuracy results of the classifiers suggest that it may be feasible to automatically classify PwD.