Abstract
INTRODUCTION: Dementia is a progressive condition that impacts cognitive and motor functions, with early symptoms often subtle and difficult to detect. Early detection is crucial for effective intervention and improved care. Recent advances in sensor technology enable continuous monitoring of human motion, providing valuable indicators of dementia and cognitive decline. METHODS: This systematic review is the first to focus exclusively on motion-based dementia detection, excluding other neurological conditions. The study aimed to address gaps in the literature by analysing evidence for motion assessment as a tool for dementia detection and by identifying and comparing sensor types, sensor placements, motion assessment tasks, extracted motion features, and analytical methods. Electronic databases (PubMed, Web of Science, IEEE Xplore and Scopus) were searched for articles published between January 2015 to May 2025. RESULTS: A total of 23 published articles were included. Sensors used across studies included inertial measurement units, pressure mats, cameras, and passive infrared sensors, with placements on body parts, wall-mounted, or floor-based. Motion assessment tasks were grouped into three categories: gait, activities of daily living, and standing postural control. Regarding analytical approaches, 11 studies employed machine learning techniques, while 12 studies utilised statistical analysis. The findings indicate that motion-based assessments demonstrate strong potential for dementia detection, as motion-related features extracted from specific tasks can serve as sensitive indicators of dementia-related cognitive decline. DISCUSSION: Compared with traditional dementia diagnostic pathways that often involve lengthy assessment cycles, this review's findings provide guidance on refining motion-based sensor selection, task design, and analytical methods to improve standardisation and reproducibility. Future research should prioritise: (1) large-scale, longitudinal data collection with confirmed dementia diagnoses to support machine learning model development; (2) standardisation of sensor types, placements, and motion metrics to enhance comparability; and (3) integration of multimodal data, including motion and brain signals, using explainable machine learning techniques to improve detection accuracy and clinical interpretability.