Optimization of velocity-storage model using human data: a computational modeling study

Objectives: The present study attempted to refine the velocity-storage (VS) model by tuning its parameters to match published human clinical data on vestibulo-ocular reflex (VOR) responses elicited by rotational and tilt paradigms.Methods: Based on the known VS framework, we stepwise tuned the parameters related to (1) the velocity-storage integrator, (2) rotational and somatogravic feedback, and (3) the linear-velocity integrator. Targets for optimization were defined as the temporal course of the VOR after earth-vertical axis rotation, with or without post-rotatory tilt, and during eccentric (“barbecue-spit”) rotation. Parameter fitting in MATLAB (MathWorks) minimized discrepancies between simulated vestibular signals and published human VOR responses.Results: Optimization yielded a velocity-storage feedforward gain of 0.12 and a leak of 0.092 per second (time constant 10.9 seconds). Rotational and somatogravic feedback gains were optimized to 0.16 and 0.76, respectively. With a fixation distance of 1 m, linear-velocity integrator optimization produced an overall gain of 0.60 and a leak of 12.5 per second (time constant 0.08 seconds). The refined model reproduced human VOR responses in the post-rotatory tilt paradigm, yielding decay time constants of 5.4 seconds in nose-down and 3.7 seconds in right-ear-down tilts, where angular VOR and translational VOR act antagonistically, and 7.2 seconds in nose-up and 5.9 seconds in left-ear-down tilts, where they act synergistically. In earth-horizontal “barbecue-spit” rotation, simulations showed sinusoidal modulation plus a steady bias larger than typical human values.Conclusions: The tuned model may be applied to predict human vestibular responses during everyday movements, and it can be extended to characterize certain pathologic manifestations.