Kalman filter detection of blinks in video-oculography: applications for VVOR measurement during locomotion

A Kalman filter algorithm was implemented for automatic detection of blink artifacts in video-oculography (VOG) data, and a cubic spline used to patch the eliminated data. The algorithm was tested by randomly introducing artificial blinks into eye movement data and computing the errors introduced by the patches. We also computed visual vestibulo-ocular reflex (VVOR) gain and phase in healthy and vestibulopathic subjects during a locomotor task, before and after blink removal, to demonstrate the interpretive importance of eliminating blink artifacts. The error introduced by the patched data was small (0.50+/-0.32 degrees ) and within the resolution of head angle measurements. Comparison of gain and phase shift before and after removing blinks revealed that even when calculated values are within expected limits, coherence of the VVOR signal was significantly (p=0.003) lower prior to blink removal (0.51+/-0.37) compared to that after blink removal (0.92+/-0.08). Comparison of VVOR calculations between healthy and vestibulopathic subjects (after removal of blinks) revealed that vestibulopathic subjects had significantly decreased gains (p=0.018) and increased phase shifts (p=0.009): these results agree with data reported in literature. We conclude that the Kalman filter detection and cubic spline patching algorithms are useful tools for VOG and should enable reliable VVOR measurements during unconstrained, ecologically meaningful locomotor activities.