Discriminative Model for Identifying Motion Primitives Based on Virtual Reality-Based IADL

Although the number of patients with dementia has been increasing with the aging of society, no efficient treatment for advanced dementia forms has been proposed yet. Therefore, it is important to detect mild cognitive impairment (MCI) to prevent dementia from further progressing. According to a recent study, people with MCI tend to perform more inefficiently compared with healthy older adults during performance-based tests on instrumental activities of daily living (IADL). In this research, we aim to develop a discriminative model to identify motion primitives based on virtual reality-based IADL. In the experiment, finger movement was measured through the Lunchbox Task simulating meal preparation using a touch panel and a three-dimensional motion sensor. The time series velocity was estimated from the obtained position data, and segmentation was conducted based on the property of the reaching motion between two points. Considering the data of each segment interval, feature extraction and coding were performed according to the predefined motion primitives, and modeling based on machine learning was implemented. As a result, the identification accuracy of motion primitives was 97.1%. Sensitivity by category was 99% for stationary actions, 93% for pointing, 97% for drag, and 93% for click or release. These results indicate that hand movements during VR-IADL can be classified into four categories based on their behavioral characteristics. On this basis, the automatic identification of the motion primitive in VR-IADL is deemed realizable.