Head-dynamic admissibility: physically constrained separation of estimated head motion and sensor-relative motion in wearable inertial recordings
Year: 2026
Type: preprint
Abstract: Background. A wearable inertial sensor measures the movement of its own housing. Even when the device sits firmly against the mastoid, a recorded head-impact waveform can contain movement of the sensor with respect to the skull. Low-pass filtering and whole-record rejection suppress unwanted structure, but neither estimates which part of a measured angular-velocity trajectory is compatible with movement of the head. Methods. We developed head-dynamic admissibility (HDA), a constrained decomposition of triaxial gyroscope signals into an estimated head angular velocity and a bounded sensor-relative component. The model limits sensor-relative angular excursion and minimises post-contact inertial demand, and it accounts for gyroscope quantisation and saturation. We tested recovery on synthetic trajectories carrying known injected relative movement, then applied HDA version 0.1.2 to 341 video-classified football events (233 headers and 108 non-headers) and to 17,607 gyroscope-analyzable recordings from two women's elite ice-hockey teams during the 2025–2026 season. We calculated frozen Impact Feature Space (IFS) descriptors and H1 scores from the recorded signal and from the HDA head estimate. Results. Synthetic injection tests separated injected sensor-relative ringing from smooth-head nulls with an area under the receiver operating characteristic curve of 0.94 to 1.00 across the tested configurations; waveform correlations were 0.83 to 0.91 and head-trajectory error fell by 43 to 58%. In both field datasets the size of the correction depended on the class of descriptor. Integral angular activity changed little in the video-classified football recordings (median 2.57% in headers and 2.26% in non-headers), whereas C01, C03, C06 and C07 changed substantially. The same pattern held across the 17,607 hockey recordings, where median changes were 1.69% for C05 and 4.51% for C02 against 25.91% for C01, 25.90% for C03 and 21.13% for C07. Overall H1 changed by a median of − 2.73 points. Conclusions. From a single mastoid-mounted gyroscope, HDA yields an explicitly constrained estimate of head angular motion together with the sensor-relative component that accompanies it. Because the size of that correction varies systematically across waveform descriptors, sensor coupling can be assessed directly when single events are interpreted and when longitudinal mechanical-exposure summaries are built. The field recordings carry no independent skull kinematics against which the estimate could be checked.
Source: Research Square
Institution: University of Helsinki
Language: English
Subfield: Clinical Neurology
Field: Medicine
Domain: Health Sciences
Open Access status: green
OpenAlex ID: W7236824790
