Classification of dementia type using the brain-computer interface

Abstract This paper addresses the development of a dementia screening tool using a character-input-type brain–computer Interface (BCI). A blinking letter board is presented to the subject for each matrix by the character-input-type BCI, and by keeping an eye on one character, the character-gazing is estimated based on the event-related potential P300 of the subject. In this experiment, the subject is instructed to specify and subsequently watch a task character. Four sets are made, each consisting of five or six task letters per subject. The subjects include 53 elderly people in their 60 s and 90 s who were diagnosed with specific symptoms of dementia. The dementia types of the subjects include the Alzheimer’s type of dementia (AD), the Lewy body type of dementia, as well as the mild cognitive impairment (MCI). The relationship between the types of dementia and the four BCI features is explained by the Kruskal–Wallis test and multiple comparisons. Also, dementia types are classified using the BCI features that are closely related to each specific type. The results were obtained using four BCI features as inputs to the classifier and three dementia types as outputs. The classification rate for the three groups was about 60%. Since the classification rate of dementia with the Lewy body (DLB) is low, the classification was performed in two groups, MCI and AD. Furthermore, the classification rate of about 80% was confirmed.