Exploring within-session separability between video-annotated self-injurious behavior and accelerometer-derived features in individuals with severe intellectual disabilities
Background Wearable accelerometry may support monitoring of movement-based self-injurious behavior (SIB) in individuals with autism spectrum disorder and severe intellectual disabilities. Before applying automated detection models, it is important to examine whether video-annotated SIB events are distinguishable from non-SIB activity within the same observation context. Objective This brief report explored correspondence between video-annotated SIB events and accelerometer-derived features in two individuals with severe intellectual disabilities, focusing on within-session separability and exploratory threshold tradeoffs. Methods Triaxial acceleration was recorded using wrist-worn sensors during video-recorded consultation sessions. SIB events were annotated from video and aligned with accelerometer data. Multiple features, including vector magnitude, median-centered vector magnitude, first-difference features, jerk, and rolling-window variability features, were extracted. For each feature, SIB event peaks were compared with the upper tail of non-SIB activity within the same observation window. Exploratory threshold sweeps characterized tradeoffs between capturing annotated events and producing false-positive candidate segments. Results The two cases showed contrasting patterns. In Case A, SIB events were relatively well separated from non-SIB activity, and threshold exploration identified ranges that retained most annotated events while producing few false-positive candidate segments. In Case B, SIB event peaks substantially overlapped with high-acceleration non-SIB activity. Capturing all annotated events required highly permissive thresholds, whereas thresholds limiting false-positive candidates missed all annotated events. Conclusion Accelerometer-based SIB monitoring should be interpreted in relation to within-person, within-window separability between target events and non-SIB activity. Video-anchored feature exploration may be a useful preliminary step before developing automated or semi-automated SIB monitoring systems.
Authors
- 健嗣 鈴木
- Atsushi Harada (ORCID: https://orcid.org/0000-0002-9485-2110)
- Soichiro Matsuda (ORCID: https://orcid.org/0000-0002-9257-2502)
- Masahiko Inoue
- Keita Nakatani
Institutions
- University of Tsukuba (JP)
- National Institute of Technology, Yonago College (JP)
- Tottori University (JP)
Publication Details
- Journal
- Frontiers in Digital Health
- Published
- 2026-11-18
- DOI
- https://doi.org/10.3389/fdgth.2026.1909340
- Primary Topic
- Context-Aware Activity Recognition Systems
- Type
- article
- Field-Weighted Citation Impact
- 0.00
Funders
- Japan Agency for Medical Research and Development