Alzheimer's patient activity assessment using different sensors

International audiencePurpose: The number of people requiring care, including Alzheimer patients, will grow while the number of people able to provide care will decrease. We focus on the development of medical, information and communication technologies for improved diagnosis and evaluation of dementia progression in early-stage Alzheimer disease (AD) patients. Method: We compared several sensors (video and accelerometer) to assess elderly performance on instrumental activities of daily living (IADL) and gait tests in the clinical protocol developed and executed by the Memory Center of the Nice Hospital and the Department of Neurology at National Cheng Kung University Hospital, Taiwan. This clinical protocol defines a set of daily living tasks (e.g., preparing coffee, watching TV), and physical tests (e.g. a balance test), that could be realistic achieved in the designed observation room, and at the same time provide objective information about dementia symptoms. Previous works analyzed only accelerometer sensors for elderly gait analysis for dementia symptoms differentiation, while video-sensors were dedicated to ADL-detection. The comparison of several sensors could provide new evidence about patients' activities. The proposed systems used a constraint-based ontology to model and detect events based on the sensors data output. 2D-video stream data is converted to 3D-geometric information that is combined with a priori semantic information about the clinical scenario. The ontology language is declarative and intuitive (as it uses natural terms), allowing medical experts to easily define and modify the IADL and gait events models (using spatial, temporal, video-tracking and accelerometers data to describe events). The sensor system has been tested with 44 participants (healthy=21, AD=23). A stride detection algorithm was developed by the Taiwan team for the automatically acquisition of gait information using a triaxial accelerometer embedded in a wearable device. It acquires data about the participant locomotion (e.g., walking time, stride length, stride frequency). It was tested with 33 participants (healthy=17, AD=16), on a 40-meter walking test. Results & Discussion: This monitoring system detected the full set of activities of the first part of the clinical protocol (e.g., balance test, repeated sequence of sitting-standing positions) with a detection rate of 96.9% to 100% (true positive rate)